Classification of students' academic performance by different feature selection techniques by machine learning methods

Increasing academic success is always desired for students and educators. It is very important to know beforehand the factors affecting academic success. In the study, it is aimed to classify the academic performance of secondary school students by using different feature selection techniques and machine learning algorithms. The dataset used was obtained from 728 students in a secondary school affiliated to the Ministry of National Education and includes 24 features. Preprocessing and feature selection techniques were applied on the obtained data set. In the data preprocessing stage, the categorical values in the data set were converted into numerical values. Since the data set used has an unbalanced class distribution, it has been balanced using the SMOTE algorithm. A total of 8 data sets were obtained by using the basic data set and 7 different feature selection techniques. These datasets are split by Hold out and 5-fold cross-validation methods. Students' achievements at the end of the year were classified using 3-level classification technique and 9 different algorithms. The most suitable parameter sets for the classification algorithms used in the study were determined by trial and error method. Without using feature selection methods, the accuracy value was reached to 95.17% with the Gradient Increasing algorithm established with 23 features in the data set. With the data set reduced to 11 features by using embedded methods and feature selection methods, the accuracy value in the Gradient Increase algorithm of the establishment decreased to 93.31%, but the accuracy value increased to 95.54% in the Random Forest algorithm. As a result, the Random Forest algorithm was found to be the most effective algorithm for predicting student performance when used with embedded methods feature selection techniques.