Prediction of student success status with machine learning

In this study, five different data sets from the Kaggle platform were used to predict student success. Data set 1 includes 6607 observations and 20 variables; Data set 2 includes 10,000 observations and 15 variables; Data set 3 includes 1388 observations and 5 variables; Data set 4 includes 10,000 observations and 6 variables; Data set 5 includes 2392 observations and 15 variables. Missing and outlier analysis was performed on the data sets, and necessary corrections were made. Categorical variables were converted to numerical format by label encoding method, and then correlation analysis was performed to remove variables with low correlation. The performances of the models were evaluated via coefficient of determination (R²), mean absolute error (MAE) and root mean square error (RMSE) metrics with 10-fold cross-validation using Ridge, Linear, Lasso, Random Forest and Support Vector Regression (SVR) algorithms. The models were re-trained by applying the standardization process with the Z-score method, and hyperparameter optimization was performed with GridSearchCV. The best results were obtained on dataset 1 with Random Forest algorithm (R²: 0.9553, MAE: 0.5586, RMSE: 0.7211), on dataset 2 with DVR (R²: 0.9407, MAE: 3.6525, RMSE: 4.2676), on dataset 3 with Random Forest (R²: 0.9940, MAE: 0.5378, RMSE: 0.7358), on dataset 4 with DVR (R²: 0.9999, MAE: 0.0762, RMSE: 0.1858) and on dataset 5 with Linear Regression (R²: 1.000, MAE: 0.000, RMSE: 0.000). In the final stage, a web application was developed to observe how the model that allows the prediction of student success works on real data. This application has demonstrated its usability as a decision support mechanism in personalized education processes by providing special success predictions for students. As a result of the study, it was revealed that data preprocessing, model selection and optimization steps significantly increased the prediction accuracy and that machine learning models can predict student success with high accuracy.