Prediction of academic achievements of vocational
and technical high school (VTS) students in science
courses through artificial neural networks (comparison
of Turkey and Malaysia)
Ali Yağci1&Mustafa Çevik1
Received: 16 December 2018 / Accepted: 8 February 2019 / Published online: 5 March 2019 # Springer Science+Business Media, LLC, part of Springer Nature 2019
Abstract
This study aims to predict the academic achievements of Turkish and Malaysian vocational and technical high school (VTS) students in science courses (physics, chemistry and biology) through artificial neural networks (ANN) and to put forth the measures to be taken against their failure. The study population consisted of 10th and 11th grade 922 VTS students in Turkey and 1050 VTS students in Malaysia. The study was conducted with the screening model, and a 34-item demographic questionnaire was developed for the collection of data Using the SPSS 24.0, the KR20 reliability coefficient of the questionnaire was found to be .90. The items in the questionnaire that were believed to affect academic achievement were accepted as independent variable/input, and the academic achievement averages of students in the previous year’s physics, chemistry and biology courses were considered as dependent variables/output. Using these parameters, a model was created and the academic achievements of the students were predicted with ANN using the Matlab R2016a program. At the end of the study, a successful academic achievement prediction system was developed with an average 98.0% sensitivity over 922 samples for Turkey and with a 95.7% sensitivity over 1050 samples for Malaysia, and the measures to be taken were determined in order the prevent failure of students.
Keywords Vocational and technical high school . Artificial neural network . Science courses . Academic achievement prediction
* Ali Yağci
ayagci89@gmail.com Mustafa Çevik
mustafacevik@kmu.edu.tr
1
1 Introduction
Vocational and technical education institutions, which play an important role in the development of the country, need to increase their productivity in education in order to educate human resources in the number and quality needed by the society (Erol2010). In order to increase productivity in education, what should be asked first is whether the educational programs confirm the personality traits of the target group, job require-ments and institutional strategies and whether these factors are taken into consideration when preparing the curriculum. Therefore, performing a needs analysis is a prerequisite for an effective curriculum. Therefore, one of the most important topics in educational research is whether the needs assessment was made before the education (Aycan and Balcı2001). It is vital for the vocational high schools to make the needs assessments. This is because, vocational high school students are expected not only to complete their training that supports their cultural and personal development, but also to complete their vocational training at the best level to ensure their future (Ünal, 2001, c.f.: Eker 2007). Vocational high schools are educational institutions aiming at educating students to get work experience at an early age and to be engaged in work life courses that require more skills and practice in order to meet the need of intermediate staff needed in the society. For this reason, vocational high school graduates need to educate them-selves in every sense. These characteristics of the students should be taken into consideration while preparing the curriculum of the general culture courses, especially the science courses, in addition to the vocational courses (Çevik2014). From this point of view, the success of vocational education, which is an area that can be considered as the cornerstone of the society, is one of the factors directly affecting the socioeconomic structure of the country. In this context, what should be done for qualified student training in VTSs and what measures should be taken to prevent failures should be the main objective of educational studies. Especially the achievements of VTS students in science courses are among the key elements affecting these dynamics. The achievement average of VTSs in Turkey is below the overall average. This was also statistically observed in the university entrance exams organized by ÖSYM (Student Selection and Placement Center). For example, the achievement average of the MS (mathematics-science) courses of the vocational high schools in 2016 was below the country average (ÖSYM2016).
In recent years, the Turkish Ministry of Education (MoNE), which has recently undertaken a task to ensure that young people receive education based on their interests and skills by classifying according to vocational skills such as wood technology, metal, electrical, computer and automotive, have been unable to meet the needs of their students in science and non-math courses (Binici and Arı 2004; Şahin and Fındık 2008; Yörük et al.2002). Science courses (physics, chemistry and biology) are some of these courses and they are considered to be very important for vocational high school students to be better prepared for the profession, to improve their working life achieve-ments and to be educated in the quality required by contemporary technology (MoNE 2005). It is vital to prevent potential failures in these courses. The methods used in the literature to determine the variables considered to affect student achievement and the effect of these variables are generally regression based methods. However, one of the methods used successfully in every field that needs data mining is the ANN, whether it is natural sciences, health sciences or social sciences (Han et al2011).
1.1 Artificial neural networks (ANN) and their usage areas
ANN technology, which is the result of mathematical modeling of the learning process inspired by the neurophysical structure of the human brain within the scope of Artificial Intelligence (AI) studies, begins with the implementation of neuron structure, which are biological units in the brain, to computer systems. For this reason, ANN, which is a sub-branch of AI, forms the basis of learning systems. ANN provides learning capacity from input data in computer systems, similar to learning ability of humankind through experience. ANN is one of the basic tools used in machine learning. Since it focuses on neural part, it is inspired by the brain system, aiming at learning and acquiring knowledge. Neural networks consist of input and output layers, as well as a hidden layer that can transform the input suitable for the output layer. The ANN method aims clustering and classification of the given samples for continuous classification of the following inputs (Öztemel2003). Human brain modeling through ANN has emerged with the desire to create machines that function like the human brain in the realm of digital computer modeling (Tosun2007). ANN is a computer system that performs the learning function, which is the most basic feature of the human brain. A model is created with ANN in accordance with the inputs and outputs. Inputs are information coming from an external environment or another neurons to the artificial neuron. These inputs are composed of the samples to be learned by the neuron (Yurtoğlu2005; Baş 2006). Learning in artificial neural networks is accomplished by changing the weights between the neurons. Accordingly, networks that can dynamically change the weights on the connections between the neurons through a certain method can be trained. Learning networks can recognize new forms or decide the class of a given input. Although there are a limited number of applications in the field of education about ANN today, this method has been successfully used in fields such as production, control, transportation and aviation, medicine, engineering, meteorology, biomedical and pharmaceutical industry, finance, stock market and credit card applications. It is seen that ANN and regression analysis are used together in the educational literature especially in academic achievement estimates.
1.2 Educational studies conducted using artificial neural networks in the world One of the first studies in which the ANN was used in the field of education was the comparison of multi-linear regression and stepwise linear regression analysis and artificial neural network analysis conducted by Gorr et al. (1994), for predicting the academic achievement and GPAs of the students. Subbanarasimha et al. (2000) used two different datasets to compare artificial neural networks and regression techniques in estimating student academic achievement. As a result of the study, it was found that the analysis made using artificial neural networks has a higher percentage of correct prediction. Ibrahim and Rusli (2007) have compared artificial neural networks, decision trees and linear regression methods in student achievement prediction. As a result of the research, ANN analysis was found to give better results than the others. In their study, Oladokun et al. (2008) aimed to test artificial neural network analysis in determining the variables that influenced student achievement and in estimating student perfor-mance. At the end of their study, they have concluded that ANN has a high predictive power and it is effective in determining variables that affect achievement. In their study
that uses ANN and logistic regression analysis for the classification of student achieve-ment in higher education, Çırak and Çokluk (2013) have achieved 66.10% correct classification rate with logistic regression analysis and 70.16% correct classification rate with the ANN model. As seen in the literature in recent years, there is a rapid increase in the studies that use ANN especially in the academic achievement estimates. In their study, Aryadoust and Baghaei (2016) have investigated the relationship between reading comprehension, oral and written knowledge of foreign language students in English using ANN. At the same time, the study confirms previous research in the sense that vocabulary is more important than grammar knowledge for reading comprehension. In their study, Briggs and Circi (2017) suggested ANN as a promising approach for the classification of students’ psychological quality hierarchy into different levels. In their study, Yinr Holmes et al. (2018) reported a study, in which 44 undergraduate students answered multiple-choice questions about computer pro-gramming. ANN has been used to classify nonverbal behaviors, data-response scores, and real-time comprehension statuses. In his study, Bahadır (2016) presented the results of the LRA and ANN empirical benchmarking study conducted to estimate the academic achievement of prospective mathematics teachers. In his study, Aydoğdu (2017) aimed to determine organizational factors affecting student achievement during the academic year and to predict their success at the end of the year using the data collected during the academic season in question. In the study, one prediction model have been developed by using the ANN method with 18 variables, including 1 output and 17 input variables, and two influence-models have been developed by the LRA method, in which 24 variables were included, of which 1 dependent (prediction) and 23 independent (predictor) variables. According to the results obtained, it has been seen that success status, classified as successful/unsuccessful, using the observed and predicted measures have been significantly similar. In their study, Çavdur et al. (2018) reported an approach based on clustering and target programming to create a balanced examination program in their study. The study aims to create a balanced examination program that would satisfy the students and instructors at a certain level in terms of personal workload. VeeraManickam et al. (2018), on their study presents the student performance prediction model by proposing the Map-reduce architecture based cumulative dragonfly based neural network (CDF-NN).
Along with the advancing technology and globalization, great and rapid changes occur in every field. Educational standards are improved to keep pace with these innovations and changes brought by the twenty-first century. In this age we live in, it is very important for the individuals to be successful in all areas of life. Especially the achievement of science education has a great contribution in the progress of countries in economic and technical fields. In terms of the importance of science education provided at schools, the success of science education (Physics, Chemistry and Biology) is quite importance starting from the very first steps of VTS education. This is because the success or failure of students in science classes before and after branching out will be an indicator of their professional competence. Therefore, estimates of the achieve-ments of VTS students in Turkey are among the factors that could significantly affect country’s future. Similarly, knowing the rationale behind the academic achievements of the students studying at VTSs in the fields of technology, materials production and technical education in the developed countries of the West and South Asia will guide us in this regard. The level of development in the field of technology is directly related to
the quality of technical education. Therefore, the success of science education (Physics, Chemistry and Biology) is quite important starting from the first steps of technical education. The importance of scientific success estimates is an accepted fact all over the world in order to look at the future more firmly, to realize the risks encountered in life, to cope with the problems faced and to survive. One of the last products of humankind as an effort to discover and explore the nature in line with the technological advances, ANN is the primary method used in estimating academic achievement. In this regard, the aim of the study is to determine and compare the science course (physics, chemistry and biology) achievements of the students studying at VTSs in Turkey and Malaysia, one of the most developed countries of South Asia, using artificial neural network success and to identify the necessary precautions in order to prevent student failures. In this context, the study aims to seek answers to these research questions:
1. What is the classification accuracy rate of the model obtained through ANN analysis in line with the parameters believed to affect academic achievement in physics course of the students studying at VTSs in Turkey and Malaysia? 2. What is the classification accuracy rate of the model obtained through ANN
analysis in line with the parameters believed to affect academic achievement in chemistry course of the students studying at VTSs in Turkey and Malaysia? 3. What is the classification accuracy rate of the model obtained through ANN
analysis in line with the parameters believed to affect academic achievement in biology course of the students studying at VTSs in Turkey and Malaysia? 4. And, what are the most significant factors affecting the academic achievement of
VTS students in Turkey and Malaysia in science courses (physics, chemistry and biology) and the measures to be taken in order to prevent their failures.
2 Research method
2.1 Research model and participants
The research was structured in a screening model based on quantitative research design. Screening model is a research approach aimed at describing a present or past situation as is. In this model, research subjects or objects are tried to be defined as is, in their own conditions (Karasar2005).
The study group consisted of a total of 1972 students, 922 of whom were studying in vocational and technical high schools at 10th and 11th grade in Karaman in Turkey and 1050 of whom were studying in vocational and high schools in Johor Bahru in Malaysia in the 2016–2017 academic year. One-to-one surveys were conducted with the students. Descriptive information about the students who participated in the study is given in Table1and Table2.
Of the participants, 442 (47.9%) were females, 480 (52.1%) were males. 810 (87.9%) participants stated that they had siblings and 112 (12.1%) participants stated that they had no siblings. 878 (95.2%) participants graduated from state primary schools and 44 (4.8%) participants graduated from private primary schools. Surveys conducted with 922 students in Karaman were all applied in State Vocational and Technical High School due to lack of Private Vocational
and Technical High School. Considering mother educational background, it is seen that 53 (5.8%) had no formal education, 708 (76.8%) of them graduated from primary schools, 124 (13.4%) of them graduated from high schools and 37 (4%) of them graduated from universities. Considering father educational back-ground, it is seen that 55 (5.9%) of them had no formal education, 584 (63.3%) of them graduated from primary schools, 214 (23.2%) of them graduated from high schools and 69 (7.5%) of them graduated from universities. 730 (79.2%) students stated their families interested in the course and 191 (20.8%) of them stated that their families had no interest in the course.
Of the participants, 481 (45.8%) were females, 569 (54.2%) were males. 1002 (95.4%) participants stated that they had siblings and 48 (4.6%) participants stated that they had no siblings. 1006 (95.8%) participants graduated from state primary schools and 44 (4.2%) participants graduated from private primary schools. In the study carried out in Johor Bahr, 1025 (97.7%) were applied in the State Vocational and Technical High Schools and 25 (2.3%) were applied in
Table 1 Demographic characteristics of Turkish students
Variable Group F %
Gender Female 442 47.9
Male 480 52.1
Total 922 100
Do you have siblings? Yes 810 87.9
No 112 12.1
Total 922 100
Graduated primary school type State 878 95.2
Private 44 4.8
Total 922 100
Current school type State 922 100
Private 0 0
Total 922 100
Mother Educational Background No formal education 53 5.8
Primary school 708 76.8
High school 124 13.4
University 37 4.0
Total 922 100
Father Educational Background No formal education 55 5.9
Primary school 584 63.3
High school 214 23.2
University 69 7.5
Total 922 100
Family Interest in the Course Yes 730 79.2
No 191 20.8
the Private Vocational and Technical High Schools. Considering mother educa-tional background, it is seen that 32 (3.1%) had no formal education, 31 (2.9%) of them graduated from primary schools, 550 (52.4%) of them graduated from high schools and 437 (41.6%) of them graduated from universities. Considering father educational background, it is seen that 42 (4%) of them had no formal education, 36 (3.4%) of them graduated from primary schools, 493 (47%) of them graduated from high schools and 479 (45.6%) of them graduated from universities. 776 (64.4%) students stated their families interested in the course and 374 (35.6%) of them stated that their families had no interest in the course. 2.2 Data collection tool
In this study conducted in the screening model, an items pool was developed for the collection of data. A questionnaire was developed, consisting of 76 items selected by researchers from this pool. In developing the study
Table 2 Demographic characteristics of Malay students
Variable Group f %
Gender Female 481 45.8
Male 569 54.2
Total 1050 100
Do you have siblings? Yes 1002 95.4
No 48 4.6
Total 1050 100
Graduated primary school type State 1006 95.8
Private 44 4.2
Total 1050 100
Current school type State 1025 97.6
Private 25 2.4
Total 1050 100
Mother educational background No formal education 32 3.1
Primary school 31 2.9
High school 550 52.4
University 437 41.6
Total 1050 100
Father educational background No formal education 42 4.0
Primary school 36 3.4
High school 493 47.0
Univeristy 479 45.6
Total 1050 100
Family interest in the course Yes 776 64.4
No 374 35.6
questionnaire, opinions of 3 experts from the science field, 1 evaluation expert, 2 technical education experts were obtained and 6 items were removed due to failure to meet language and measurement criteria, and a 70-item questionnaire was obtained as a result. Thus, the content validity of the questionnaire was ensured. For the construct validity of the questionnaire, data were collected with the pilot study (250 people), and the values KMO = .86, Bartlett’s test = .00, and Chi-square = 2.183 were found, and them item analysis was performed accordingly. Items 9, 29, 33 and 34, which have factor loadings below .30, were removed from the questionnaire for item factor analysis. In order to determine the reliability, the KR20 reliability coefficient was calculated and found to be .90. The questionnaire consists of 30 items and three sub-scales. The parameters considered to affect academic achievement are grouped under the following headings:1. Student Factors, 2. Teacher-School Factors, 3. Paren-tal Factors.
2.3 Theoretical framework
Gender, self-confidence in science and parental support (Kaya and Rice2010), stu-dent’s attitude towards science (Freedman1997), student’s interest in the course (Dee 2007; Klem and Connell2004).
Te a c h e r s u p p o r t ( K l e m a n d C o n n e l l 2 0 0 4) , t e a c h e r c o m p e t e n c e (Konstantopoulos 2006), science teaching method, curriculum preparation, and suggesting research theme (Jackson and Ash2012). Teacher-student interactions, the effectiveness of the approach. Teacher characteristics, teaching variables and classroom composition (Kaya and Rice2010), class participation and cooperative learning in class (Johnson et al. 2007). School features such as school locality and characteristics, school socioeconomic status, daily attendance, college pre-paratory students, and high school graduates are important predictors for the average student (Konstantopoulos 2006).
Parental support (Kaya and Rice2010). 2.4 Application procedure
The procedure and methods as well as stages for the implementation followed in the study as follows:
1. Within the scope of CHE Project Based Mevlana Exchange Program, it has been decided to contact the instructors of Faculty of Education of Malaysia Technology University (UTM) and a decision has been reached regarding the subject of the study.
2. A work schedule was formed after the topic was determined.
3. Within the scope of the study, the topics that might affect the academic success of science courses of vocational and technical high school students were identified based on the literature.
4. The 34 factors identified were transformed into a yes-no choice structure (76 questions were obtained). For the validity of the survey, expert opinions were taken in the field of measurement, science and informatics.
5. The survey were applied rigorously to 922 students studying in Grade 10 and 11 in VTHS in Turkey
6. This developed survey was translated into Malay to be applied in Malaysia and the final controls were carried out by 3 experts in technical education, measurement and information technology disciplines..
7. The survey was applied to 1050 students studying in 10th and 11th grades at VTHS in Malaysia.
8. Data obtained from both countries were entered by the researcher into the ANN program coded as 0 and 1 and the estimation percentages were tried to be reached.
2.5 Data analysis
ANN analysis was used in the statistical analysis process. The GPAs of the students at the end of the year were determined as output variables for ANN in the study. Outputs were arranged in accordance with the system of 5, which is a passing grade system for students studying in Turkey determined by Ministry of Education (MoNE2016). The data are given in Table3.
As seen in Table3, for those with a score of 50 points or more, the input of the ANN was coded as 1, whereas for those with 49 points and less the input of the ANN was coded as 0. Education System in Malaysia (ESM) (2019), which was determined by the Malaysian Ministry of Education, was taken as the basis in the previous year science course academic achievement averages (Table4). According to this;
As seen in Table4, according to the Malaysian Education system the course passing grade in which scores between 90 and 100 are identified as A+, between 80 and 89 identified as A, between 70 and 79 identified as A-, between 65 and 69 identified as B+ , between 60 and 64 identified B, between 55 and 59 identified as C+, between 50 and 54 identified as C, between 45 and 49 identified as D, between 40 and 44 identified as E and between 0 and 39 identified as G was considered. Scores between 1 and 55 were taken asBsuccessful^ and scores between 0 and 54 as were taken as Bunsuccessful^. The items in the questionnaire that were believed to affect academic achievement were accepted as input. Using inputs and outputs, a model was created and the academic achievement of the students was predicted with ANN in the Matlab R2016a program. The learning rule-set Scaled Conjugate Gradient (TrainsCG) algorithm was used in the training of the feed-forward back-propagation network. The learning rate (which determines the learning speed of ANN) is 0.01 and the coefficient of momentum (a factor that helps the network recover faster) is 0.9 (Sabancı2013). A three-layer feed-Table 3 Course success scores in
the Turkish education system and input into ANN
ANN code Score Mark
0 85–100 5 70–84 4 55–69 3 50–54 2 25–49 1 0–24 0
forward network was used, which are input, hidden and output layers, in the study. Sigmoid function was used in the hidden layer and linear activation function was used in the output layer. For Turkey, 80% (738) of the data set was used for training, 5% (46) was used for validation, and 15% (138) was used for testing, and for Malaysia, 80% (839) of the data set was used for training, 5% (53) was used for validation, and 15% (158) was used for testing (Fig.1).
Figure 1 shows ratios in the ANN architecture for Turkey. These settings are defaulted on the system side. According to the results, these ratios are changed and
Table 4 Course success scores in the malay education system and input into ANN
ANN code Score Letter Grade
1 90–100 A+ 80–89 A 70–79 A-65–69 B+ 60–64 B 55–59 C+ 0 50–54 C 45–49 D 40–44 E 0–39 G
the optimum result is approached. For Turkey, 80% (839) of the data set was used in training, 5% (53) of the data set was used in validation and 15% (158) of the data set was used in the testing. The ANN used in the study has 30 layers, 70 inputs and 1 output. It is shown in Fig.2.
Figure 2 shows ratios in the ANN architecture for Malesia. These settings are defaulted on the system side. According to the results, these ratios are changed and the optimum result is approached. The ANN used in the study has 30 layers, 70 inputs and 1 output. The algorithm of ANN model, which was designed for success estimation performed by he Neural Pattern Recognition (nprtool) command in MATLAB2016a, was given in Fig.3.
Fig. 2 Training, validation and test ratios in the system designed for Malaysia
Fig. 3 The ANN architecture for each course academic achievement prediction with MATLAB2016a. Number of Artificial Neural Network Input Layers:70. Neural Network Output Layer Number:1. Sample Number: Turkey- 922, Malaysia−1050. Latent Layer Number: 30. Learning Rule: Scaled Conjugate Gradient (trainscg)
3 Findings
In this section, the findings and interpretations obtained by analyzing the data in line with the purposes of the research are discussed.
3.1 ANN analysis results on prediction of physics course success of VTS students in Malaysia and Turkey
The findings of the analysis for the classification of students with ANN according to their achievement in physics course are presented in Table5.
When we look at the Table5, of the 730 individuals participating in training for physics course in Turkey, 363 of the 364 successful students were correctly classified, one of them was misclassified, and the correct classification rate was 99.7%. Of the 369 unsuccessful students, 367 were classified correctly, 2 were misclassified, and the correct classification rate was 99.5%. The overall correct classification rate was 99.6%. Of the 46 individuals participating in the validation analysis, 16 of the 17 successful students were correctly classified, 1 of them was misclassified, and the correct classification rate was 94.1%. Of the 29 unsuccessful students, 29 were classified correctly, none of them was misclassified, and the correct classification rate was 100%. The overall classification rate was 97.8%. Of the 138 individuals partici-pating in the test, 47 of the 63 successful students were correctly classified, 16 were misclassified, and the correct classification rate was 74.6%. Of the 74 unsuccessful students, 64 were classified correctly, 10 were misclassified, and the correct classifica-tion rate was 86.5%. The overall correct classificaclassifica-tion rate was 96.7%. Of the 839 individuals participating in the training in Malaysia, 42 of the 42 successful students were correctly classified, none of them was misclassified, and the correct classification
Table 5 Classification table obtained by ANN model for physics course
Turkey Malaysia
ANN Real status Estimated CEP Estimated CEP
True False True False
f % f % (%) f % f % (%) Education Successful 363 49.5 1 0.1 99.7 42 5 0 100 100 Unsuccessful 367 50.1 2 0.3 99.5 797 95 0 100 100 Total 730 99.6 3 0.4 99.6 839 100 0 100 100 Verification Successful 16 34.8 1 2.2 100 1 1.9 0 100 100 Unsuccessful 29 63.0 0 0.0 94.1 52 98.1 0 100 100 Total 45 97.8 1 2.2 97.8 53 100 0 100 100 Test Successful 47 34.3 16 11.7 86.5 1 1.3 2 0.6 33.3 Unsuccessful 64 46.7 10 7.3 74.6 147 93 8 5.1 66.7 Total 111 81.0 26 19.0 81.0 158 94.3 10 5.7 100 CEP Correct Estimation Percentage
rate was 100%. Of the 797 unsuccessful students, 797 were classified correctly, none of them was misclassified, and the correct classification rate was 100%. The overall correct classification rate was 100%. Of the 53 individuals participating in the valida-tion analysis, 1 of the 1 successful students were correctly classified, none of them was misclassified, and the correct classification rate was 100%. Of the 52 unsuccessful students, 52 were classified correctly, none of them was misclassified, and the correct classification rate was 100%. The overall correct classification rate was 100%. Of the 158 individuals participating in the test, 1 of the 3 successful students were correctly classified, 2 were misclassified, and the correct classification rate was 33.3%. Of the 155 unsuccessful students, 147 were classified correctly, 8 were misclassified, and the correct classification rate was 94.8%. The overall correct classification rate was 100%. 3.2 ANN analysis results on prediction of chemistry course success of VTS students in Malaysia and Turkey
The findings of the analysis for the classification of students with ANN according to their achievement in chemistry course are presented in Table6.
When we look at the Table6, of the 732 individuals participating in the training for chemistry course in Turkey, 423 of the 423 successful students were correctly classified, none of them was misclassified, and the correct classification rate was 100%. Of the 309 unsuccessful students, 309 were classified correctly, none of them was misclassified, and the correct classification rate was 100%. The overall correct classification rate was 100%. Of the 36 individuals participating in the validation analysis, 19 of the 19 successful students were correctly classified, none of them was misclassified, and the correct classification rate was 100%. Of the 27 unsuccessful students, 27 were classified correctly, none of them was misclassified, and the correct classification rate was 100%.
Table 6 Classification table obtained by ANN model for chemistry course
Turkey Malaysia
ANN Real status Estimated CEP Estimated CEP
True False True False
f % f % (%) f % f % (%) Education Successful 423 57.8 0 0.0 100 59 7 0 100 100 Unsuccessful 309 42.2 0 0.0 100 780 93 0 100 100 Total 732 100 0 0.0 100 839 100 0 100 100 Verification Successful 27 58.7 0 0.0 100 2 3.8 0 100 100 Unsuccessful 19 41.3 0 0.0 100 51 96.2 0 100 100 Total 46 100 0 0.0 100 53 100 0 100 100 Test Successful 83 60.1 5 3.6 94.3 0 0.0 2 1.3 0.0 Unsuccessful 45 32.6 5 3.6 90.0 145 91.8 11 7.0 92.9 Total 123 92.8 10 7.2 92.8 145 94.3 10 5.7 100 CEP Correct Estimation Percentage
The overall correct classification rate was 100%. Of the 138 individuals participating in the test, 83 of the 88 successful students were correctly classified, 5 were misclassified, and the correct classification rate was 94.3%. Of the 50 unsuccessful students, 45 were classified correctly, 5 were misclassified, and the correct classification rate was 92.9%. The overall correct classification rate was 98.9%. Of the 839 individuals participating in the training in Malaysia, 59 of the 59 successful students were correctly classified, none of them was misclassified, and the correct classification rate was 100%. Of the 780 unsuccessful students, 780 were classified correctly, none of them was misclassified, and the correct classification rate was 100%. The overall correct classification rate was 100%. Of the 53 individuals participating in the validation analysis, 2 of the 2 successful students were correctly classified, none of them was misclassified, and the correct classification rate was 100%. Of the 51 unsuccessful students, 51 were classified correctly, none of them was misclassified, and the correct classification rate was 100%. The overall correct classification rate was 100%. Of the 158 individuals partic-ipating in the test, none of the 2 successful students were classified correctly, 2 were misclassified, and the correct classification rate was 0%. Of the 156 unsuccessful students, 145 were classified correctly, 11 were misclassified, and the correct classifi-cation rate was 92.9%. The overall correct classificlassifi-cation rate was 100%.
3.3 ANN analysis results on prediction of biology course success of VTS students in Malaysia and Turkey
The findings of the analysis for the classification of students with ANN according to their achievement in biology course are presented in Table7.
When we look at the Table7, of the 733 individuals participating in the training for biology course, 439 of the 439 successful students were correctly classified, none of
Table 7 Classification table obtained by ANN model for biology course
Turkey Malaysia
ANN Real status Estimated CEP Estimated CEP
True False True False
f % f % (%) f % f % (%) Education Successful 439 59.9 0 0.0 100 104 12.4 12 1.4 89.7 Unsuccessful 294 40.1 0 0.0 100 675 80.5 48 5.7 93.4 Total 733 100 0 0.0 100 839 92.9 60 7.1 92.8 Verification Successful 27 58.7 0 0.0 100 3 5.7 0 0.0 100 Unsuccessful 19 41.3 0 0.0 100 42 79.2 8 15.1 84.0 Total 46 100 0 0.0 100 53 84.9 0 15.1 84.9 Test Successful 71 51.8 2 1.5 99.6 8 5.1 19 12.0 29.6 Unsuccessful 60 43.8 4 2.9 98.9 108 68.4 23 14.6 82.4 Total 131 95.6 42 4.4 99.3 116 73.5 42 16.6 73.4 CEP Correct Estimation Percentage
them was misclassified, and the correct classification rate was 100%. Of the 294 unsuccessful students, 294 were classified correctly, none of them was misclassified, and the correct classification rate was 100%. The overall correct classification rate was 100%. Of the 46 individuals participating in the validation analysis, 27 of the 27 successful students were correctly classified, none of them was misclassified, and the correct classification rate was 100%. Of the 19 unsuccessful students, 19 were classi-fied correctly, none of them was misclassiclassi-fied, and the correct classification rate was 100%. The overall correct classification rate was 100%. Of the 131 individuals participating in the test, 71 of the 73 successful students were correctly classified, 2 were misclassified, and the correct classification rate was 99.6%. Of the 64 unsuccess-ful students, 60 were classified correctly, 4 were misclassified, and the correct classi-fication rate was 98.9%. The overall correct classiclassi-fication rate was 99.3%. Of the 839 individuals participating in the training in Malaysia, 104 of the 116 successful students were correctly classified, 12 were misclassified, and the correct classification rate was 89.7%. Of the 723 unsuccessful students, 675 were classified correctly, 48 were misclassified, and the correct classification rate was 93.4%. The overall correct classi-fication rate was 92.8%. Of the 53 individuals participating in the validation analysis, 3 of the 3 successful students were correctly classified, none of them was misclassified, and the correct classification rate was 100%. Of the 50 unsuccessful students, 42 were classified correctly, 8 were misclassified, and the correct classification rate was 84%. The overall correct classification rate was 84.9%. Of the 158 individuals participating in the test, 8 of the 27 successful students were correctly classified, 19 were misclassified, and the correct classification rate was 29.6%. Of the 131 unsuccessful students, 108 were classified correctly, 23 were misclassified, and the correct classifi-cation rate was 82.4%. The overall correct classificlassifi-cation rate was 73.4%.
As seen in Table8, correct estimate percentages of the VTS students for the overall science courses in Turkey were 96% for physics, 98.9% for chemistry and 99.1% for
Table 8 Classification table obtained by ANN model for overall research
Turkey Malaysia
Courses Real status Estimated CEP Estimated CEP
f (%) f (%) Physics Successful 426 95.9 44 95.7 Unsuccessful 460 97.5 996 99.2 Total 886 96.0 1000 99 Chemistry Successful 533 99.1 61 96.8 Unsuccessful 373 98.7 976 98.9 Total 906 98.9 1037 98.8 Biology Successful 437 99.3 115 78.8 Unsuccessful 371 98.9 825 91.3 Total 808 99.1 940 89.5
biology. The overall correct estimate percentages for the overall study in Malaysia were 99% for physics, 98.8% for chemistry and 89.5% for biology.
3.4 The Most significant factors affecting the academic achievement of VTS students in science courses (physics, chemistry and biology) and the measures to be taken in order to prevent their failures
The most important factors affecting the academic achievement of VTS students in physics, chemistry and biology courses, with a level of significance p < .01, are presented in Table9.
As seen in Table9, the most important factor affecting failure in the physics course in both countries was the poor economic status of the family and having no room for
Table 9 The most significant factors affecting academic achievement according to ANN
Turkey Malaysia Courses Order of importance p* NSP (%) FAF p* N S P (%) FAF
Physics 1 .03 98.2 Crowded class .03 83.3 Unfavorable economic status of the family 2 .02 95.2 Not using different
methods at class
.03 75.3 Lack of infrastructure related to physics 3 .02 95.0 Failure to participate at the
activities
.02 73.9 Separated parents 4 .02 79.7 Have no room for studying .02 73.7 Not being prepared for the
exams regularly 5 .02 74.2 Unfavorable economic
status of the family
.02 69.3 Lack of the bookshelves at home
Chemistry 1 .03 100 Crowded class .03 100 Lack of Internet
2 .02 88.9 Paternal illiteracy .03 80.3 Lack of the bookshelves at home
3 .02 82.3 Paternal employment status .02 76.5 Maternal employment status
4 .02 79.4 Lack of Internet .02 74.1 Have no room for studying 5 .02 76.8 Lack of confidence in
chemical applications
.02 73.1 Maternal Vital Status Biology 1 .03 100 Does not like biology
course
.04 99.7 Difficult biology teaching program
2 .02 92.4 Not being prepared for the exams regularly
.03 79.3 Does not like biology teacher
3 .02 83.3 Separated parents .03 72.7 Lack of personal computer 4 .02 82.9 Does not like biology
teacher
.03 71.8 Paternal vital status 5 .02 82.1 Crowded class .02 68.2 Separated parents *p < .05
NSP Normalized Significance Percentage FAF Factors Affecting Failure
study or a private bookshelf. On the other hand, the crowded classes, teacher’s use of different methodologies at class, and students failure to take an active role in the physics activities were the most important factors affecting the failure of students in Turkey. The most important factors affecting the failure of the students in physics course in Malaysia, however, were the absence of a infrastructure for physics course, separated parents and failure to prepare for exams regularly. For the chemistry course, the absence of the Internet and the fact that one of the parents is unemployed was one of the factors that negatively affect the chemistry achievement in both countries. On the other hand, the crowded classes, paternal illiteracy and lack of self-confidence in the applications in chemistry course were among the most important factors in Turkey. In Malaysia, lack of study room and bookshelf, and deceased parent were among the most effective factors. For biology course, not liking the biology teacher, and separated parents were effective in both countries, whereas not liking the biology course, not preparing the exams regularly, and crowded classes were among the most effective factors in Turkey, however difficult biology curriculum, lack of computer, and deceased parent were the most important factors in Malaysia.
4 Discussion and conclusion
As a result of the study, science course achievements of VTS students in Turkey and Malaysia were estimated using ANN and these two countries were compared on this basis. In this context, firstly, an ANN model that can make highly accurate predictions of the academic achievement was created. The accuracy percentages of the predicted future success in science courses of a new student added into the model are 96% for physics, 98.9% for chemistry and 99.1% for biology course in Turkey. The multi-layered ANN model correctly classifies 95.9% of the successful students and 97.5% of the unsuccessful students in the physics course. The fact that the prediction perfor-mance for the students who failed in the physics course is higher than the prediction performance of the successful students suggests that the model created by ANN gives better results in predicting failure of the students in particular.
The multi-layered ANN model correctly classifies 99.1% of the successful students and 98.7% of the unsuccessful students in the chemistry course. In this regard, the fact that the prediction performance for the successful students in the chemistry course is higher than the prediction performance of the unsuccessful students suggests that the ANN model gives better results in predicting the successful students in particular.
The multi-layered ANN model correctly classifies 99.3% of the successful students and 98.9% of the unsuccessful students in the biology course. According to this result, the fact that the prediction performance for the successful students in the chemistry course is higher than the prediction performance of the unsuccessful students in this course suggests that the ANN model gives better results in predicting the successful students in particular. In order to influence success in physics course positively, attention should be be paid to adjust the classroom population to a pedagogically appropriate amount, to give the lecture with different teaching methods and techniques, to increase participation in course activities, to help students to have a study room, and to improve the economic status of the families. When the factors affecting the success of the chemistry course are examined, it may be concluded that the appropriate class
population, paternal literacy, paternal employment and that the student’s regular course work affects the success positively. And, it may be concluded that loving the biology course, regular preparation for exams, parents living together, loving the biology teachers, and appropriate classroom population affect the academic achievement in biology positively. Accurate prediction rate with ANN model is 99% physics, 98% for chemistry, and 89.5% for biology courses in Malaysia. The multi-layered ANN model correctly classifies 95.7% of the successful students and 99% of the unsuccessful students in the physics course. The fact that the prediction performance for the students who failed in the physics course is higher than the prediction performance of the successful students suggests that the model created by ANN gives better results in predicting failure of the students in particular. The ANN model correctly classifies 96.8% of the successful students and 98.9% of the unsuccessful students in the chemistry course. In this regard, the fact that the prediction performance for the successful students in the chemistry course is higher than the prediction performance of the unsuccessful students suggests that an ANN model gives better results in predicting the successful students in particular. The multi-layered ANN model correctly classifies 78.8% of the successful students and 91.3% of the unsuccessful students in the biology course. According to this result, the fact that the prediction performance for the successful students in the chemistry course is higher than the prediction perfor-mance of the unsuccessful students in this course suggests that the ANN model gives better results in predicting the successful students in particular. In line with the above-mentioned results, Bahadır (2013) has also reported that his model, developed for academic achievement prediction using ANN and logistic regression analysis, has higher accuracy when predicting failure, which supports our study findings. Crowded classrooms is one of the most important factors affecting failure status of VTS students in science classes in Turkey. In this case, we can say that the classes are crowded in VTSs and this affects the achievement negatively. This result is in line with the studies of Altun and Çakan (2008). Similarly, the results of the study by Yenice et al. (2013) are in parallel with our results stating that students have inadequate self-confidence in science courses, which in turn leads to failure. This is because, the science courses require application practice in addition to theory, and students need to have self-confidence in respective courses in order to participate in these practices. Similarly, it has been concluded that the familial reasons (socioeconomic status, parents living together etc.) affect the failure. This result is in line with the results of Anıl (2009), Alomar (2006), Çiftçi and Çağlar (2014) and Taningco and Pachon (2008). Similarly, it is seen that the most important factor affecting the failures of the VTS students in science courses in Malaysia is the familial reasons. One of the most important reasons for this can be attributed to the fact that the VTS students, of which the data collected, stay in boarding schools. This can also be attributed to Malaysia’s cultural values. For example, Ong et al. (2010) have also reported that the characteristics of the families is the most important factor affecting academic failure of primary school students in Malaysia. One of the most important factors affecting failure in science courses was the lack of private bookshelves, and this result is in line with the results of the studies conducted by Erbaş (2005), Özer and Anıl (2011) and Saşmazer (2006); in addition, the lack of computers and Internet was in line with the results of the studies by Özer and Anıl (2011) and Christman and Badgett (1999). This may be indicative that the use of technology in science education in Malaysia is particularly effective. In their study,
Fuchs and Woessmann (2005) concluded that computer use of the students affects their academic achievement.
As a result of the study, the primary factor that affects the failure of the VTS students in both countries is the familial factors. Specifically, not liking the biology teacher was the factor that affects failure in the biology class in both countries. It is known that the most important factor that improves the student success is the teacher factor (Ereş and BıçakKoçak 2017). In addition, the teacher is the primary factor among the other factors that trigger success. This result is parallel to the results of other studies (Erdoğdu 2006; Memduhoğlu and Tanhan2013).
4.1 Limitations
Highly accurate predictions of the ANN model in predicting the academic achievement is one of the strengths of the study. However, this research is a specific application since it was conducted with a certain amount of data obtained from VTS students. For this reason, it will be beneficial to evaluate the results obtained by taking into account the limitations in question. The trainscg learning algorithm was used for science academic success prediction of Turkish and Malay VTHS students performed by MATLAB for test data. There can be no definite explanation as to which of the optimization algorithms would work better. The performance of the optimization algorithms may vary depending on the data structure of the problem. However, the success of the algorithm decreases as the network structure expands. It is considered that the ANN analysis technique used in the study can be used in combination with logistic regression analysis or other techniques to compare the estimates and achieve more accurate predictions.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
References
Alomar, B. O. (2006). Personal and family paths to pupil achievement. Social Behavior and Personality, 34(8), 907–922.
Altun, S. A., & Çakan, M. (2008). Factors affecting student success on exams: The case of sucessful cities on LGS/OSS exams. Elementary Education Online, 7(1), 157–173.
Anıl, D. (2009). Factors effecting science achievement of science students in programme for international students’ achievement (PISA) in Turkey. Education and Science, 34(152), 87–100.
Aryadoust, V., & Baghaei, P. (2016). Does EFL readers' lexical and grammatical knowledge predict their reading ability? Insights from a perceptron artificial neural network study. Educational Assessment, 21(2), 135–156.
Aycan, Z., & Balcı, H. (2001). Individual-and organizational-level predictors of training effectiveness in business organizations. Turkish Journal of Psychology, 16(48), 13–31.
Aydoğdu, İ. (2017). Estimation of student successes by artificial Neural Networks and comparison of efficacy of impact models by logistic regression analysis. (Unpublished master dissertation).,YüzüncüYıl University, Van, Turkey.
Bahadır, E. (2013). Prediction of student teachers' academic success with logistic regression analysis and artificial neural networks methods. (Unpublished doctoral dissertation) Marmara University, İstanbul, Turkey.
Bahadır, E. (2016). Using neural network and logistic regression analys to predict prospective mathematics teachers’ academic success upon entering graduate education. Educatıonal Scıences: Theory & Practıce, 16(3), 643–964.
Baş, N. (2006). Artificial neural networks approach and an application] (unpublished master dissertation. Mimar Sinan GüzelSanatlar Universityİstanbul, Turkey.
Binici, H., & Arı, N. (2004). Seeking new perspectives in technical and vocational education. Gazi University Education Faculty Journal, 24(3), 383–396.
Briggs, D. C., & Circi, R. (2017). Challenges to the use of artificial neural networks for diagnostic classifications with student test data. International Journal of Testing, 17(4), 302–321.
Çavdur, F., Değirmen, S., & Küçük, M. K. (2018). A clustering and goal programming-based approach for homogeneous exam distribution in exam scheduling problems. Uludağ University Journal of the Faculty of Engineering, 23(1), 167–188.
Çevik, M. (2014). The evaluation of the Current Biology currucilum according to teachers’ and students’ opinions who are studying at vocational and technical high school and a new draft proposal: the example of photosynthesis subject. (Unpublished doctoral dissertation). Ankara, Turkey: Gazi University Institute of Educational Sciences.
Christman, E., & Badgett, J. (1999). A comparative analysis of the effects of computer-assisted instruction on student achievement in differing science and demographic areas. Journal of Computers in Mathematics and Science Teaching, 18, 135–143.
Çiftçi, C., & Çağlar, A. (2014). The effect of socio-economic characteristics of parents on student achieve-ment: Ispoverty destiny? International Journal of Human Sciences, 11(2), 155–175.
Çırak, G., & Çokluk, Ö. (2013). The usage of artifical neural network and logistic regresssion methods in the classification of student achievement in higher education. Mediterranean Journal of Humanities, 3(2), 71–79.
Dee, T. S. (2007). Teachers and the gender gaps in student achievement. Journal of Human Resources, 42(3), 528–554.
E d u c a t i o n S y s t e m i n M a l a y s i a ( E S M ) . ( 2 0 1 9 ) . r e t r i v e d f r o m h t t p s : / / w w w. s c h o l a r o . com/pro/Countries/Malaysia/Grading-Systemon 05.02.2019.
Eker, G. (2007). The burnout level students of occupational high school. (Unpublished master dissertation). İstanbul, Turkey: Marmara University.
Erbaş, K. C. (2005). Factors affecting scientific literacy of students in Turkey in programme for international student assessment (PISA). (Unpublished master dissertation). Ankara, Turkey: Middle East University. Erdoğdu, M. Y. (2006). Relationships between creativity, teacher behaviours and academic success. Elektronik
Sosyal Bilimler Dergisi, 5(17), 95–106.
Ereş, F., & BıçakKoçak, D. (2017). Increasing and hindering factors of student achievement in middle schools. ASOS Journal., 5(51), 32–45.
Erol, N. (2010). The determining the educational needs of teachers in accordance with new curriculum in vocational high education. (Unpublished master dissertation). Ankara, Turkey: Gazi University. Freedman, M. P. (1997). Relationship among laboratory instruction, attitude toward science, and achievement
in science knowledge. Journal of Research in Science Teaching, 34(4), 343–357.
Fuchs, T., & Woessmann, L. (2005). Computers and student learning: bivariate and multivariate evidence on the availability and use of computers at home and at school. (CES) Ifo Working Paper Series No.8). Retrieved fromhttp://www.cesifo-group.de/DocDL/IfoWorkingPaper-8.pdf.
Gorr, W. L., Nagin, D., & Szczypula, J. (1994). Comparative study of artificial neural network and statistical models for predicting student grade point averages. Interntional Journal of Forecasting, 10, 17–34. Han, J. Pei, J., & Kamber, M. (2011). Data mining, Southeast Asia edition: concepts and techniques.
Amsterdam: Elsevier.
Holmes, M., Latham, A., Crockett, K., & O'Shea, J. D. (2018). Near real-time comprehension classification with artificial neural networks: Decoding e-learner non-verbal behavior. IEEE Transactions on Learning Technologies, 11(1), 5–12.
Ibrahim, Z., & Rusli, D. (2007). Predicting students’ academic performance: comparıng artificial neural network, decision tree and linear regression. 21st Annual SAS Malaysia Forum, Shangri- La Hotel, Kuala Lumpur, September, 5.
Jackson, J. K., & Ash, G. (2012). Science achievement for all: Improving science performance and closing achievement gaps. Journal of Science Teacher Education, 23(7), 723–744.
Johnson, C. C., Kahle, J. B., & Fargo, J. D. (2007). A study of the effect of sustained, whole-school professional development on student achievement in science. Journal of Research in Science Teaching, 44(6), 775–786.
Kaya, S., & Rice, D. C. (2010). Multilevel effects of student and classroom factors on elementary science achievement in five countries. International Journal of Science Education, 32(10), 1337–1363. Klem, A. M., & Connell, J. P. (2004). Relationships matter: Linking teacher support to student engagement
and achievement. The Journal of School Health, 74(7), 262–273.
Konstantopoulos, S. (2006). Trends of school effects on student achievement: Evidence. Teachers College Record, 108(12), 2550–2581.
Memduhoğlu, H. B., & Tanhan, F. (2013). Study of organisational factors scale’s validity and reliability affecting university students’ academic achievements. YüzüncüYıl University Journal Of Education Faculty, 10(1), 106–124.
Ministry of National Education (MoNE), (2016). Ministry of National Education Regulation of Secondary Education Institutions. Ankara, Ministry of National Education Publish. retrived fromhttp://ogm.meb. gov.tr/meb_iys_dosyalar/2016_11/03111224_ooky.pdfon 04.02.2019.
MoNE, (2005). Mathematics and vocational mathematics curriculum. Ministry of National Education General Directorate of Apprenticeship and Non-formal Education, Ankara.
Oladokun, V. O., Adebanjo, A. T., & Charles-Owaba, O. E. (2008). Predicting students’ academic perfor-mance using artificial neural network: A case study of an engineering course. The Pacific Journal of Science and Technology, 9(1), 72–79.
Ong, L. C., Chandran, V., Lim, Y. Y., Chen, A. H., & Poh, B. K. (2010). Factors associated with poor academic achievement among urban primary school children in Malaysia. Singapore Medical Journal, 51(3), 247–252.
ÖSYM, (2016). Undergraduate placement examinations (2016-Lys) Results. Retrieved fromhttp://dokuman. osym.gov.tr/pdfdokuman/2016/LYS/LYSSayisalBilgiler19072016.pdf
Özer, Y., & Anıl, D. (2011). Examining the factors affecting students' science and mathematics achievement with structural equation modeling. Hacettepe University Journal Of Education Faculty, 41, 313–324. Öztemel, E. (2003). Artificial Neural Networks (First Edition).İstanbul: Papatya Publishing.
Sabancı, K. (2013). Determination of variable-level herbicide application parameters by artificial neural networks for weed control in sugar beet cultivation. Ph.D. Thesis, Selcuk University, Institute of Science and Technology, Konya.
Şahin, İ., & Fındık, T. (2008). Vocational and technical education in Turkey: Current situation, problems and proposition solutions. TSA, 12(3), 65–86.
Saşmazer, G. A. (2006). Factors that affecting success of scientific literacy on students in Turkey that participate programme for international student assessment (PISA). (Unpublished master dissertation). Ankara: Hacettepe University.
Subbanarasimha, P. N., Arinzeb, B., & Anandarajanb, M. (2000). The predictive accuracy of artificial neural networks and multiple regression in the case of skewed data. Exploration of some issues. Expert Systems with Applications, 19, 117–123.
Taningco, M. T. V., & Pachon, H. P. (2008). Computer use, parental expectations, and latino academic achievement. USA: Tomas Rivera Policy Institute.
Tosun, Ö. (2007). Artificial neural networks decision tree comparison in classification analysis: An applica-tion on students success. (unpublished master dissertaapplica-tion). İstanbul, Turkey: İstanbul Technical University.
VeeraManickam, M. R. M., Mohanapriya, M., Pandey, B. K., Akhade, S., Kale, S. A., Patil, R., & Vigneshwar, M. (2018). Map-reduce framework based cluster architecture for academic student’s perfor-mance prediction using cumulative dragonfly based neural network. Cluster Computing, 1–17. Yenice, N., Saydam, G., & Telli, S. (2013). Determining factors effecting on primary school students’
motivation towards science learning. Ahi Evran University Kırşehir Education Faculty Journal (KEFAD), 13(2), 231–247.
Yörük, S., Dikici, A., & Uysal, A. (2002). The informati on of community and vocational education in Turkey. Fırat University Journal of Social Science, 24(3), 229–312.
Yurtoğlu, H. (2005). Artificial neural networks and predictive modeling methodology: in Turkey for some macroeconomic variables. (Published master dissertation). Ankara: State Planning Organization.
