Predicting student academic achievement using stacked ensemble learning with deep neural networks and fuzzy-based feature selection

Enhancing student performance and academic planning can be greatly impacted by predicting students' growth. It is possible to customize educational programs to each student's performance by offering comprehensive insights into their needs and weaknesses. These methods can also aid in recognizing and averting academic issues, which will ultimately improve pupils' academic performance. In this regard, we have presented a novel method that achieves these goals. The proposed method consists of four basic steps, in the first step, preprocessing of raw data is performed. In the next step, a fuzzy logic-based hybrid model is used to select features related to students' academic performance. In this method, first, each of the preprocessed features is ranked using Mutual Information (MI) and Analysis of Variance (ANOVA) measures. Then, these rankings are combined with the help of a fuzzy inference model and the features are ranked based on the rules of the fuzzy model. Finally, using the backward elimination feature selection (BEFS) technique, irrelevant features are eliminated and relevant features are selected. In the third step, modeling is performed using three deep neural networks including Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and Multilayer Perceptron (MLP), each of which independently tries to model the target variable. In the final step, a meta-model based on the MLP structure is used to extract the target variable based on the predictions from the three deep ensemble models. According to the results obtained through evaluating the model by a questionnaire-based dataset, the proposed methodology achieves significant improvements in predictive accuracy (RMSE 0.6%, MAPE 0.03%), offering a valuable tool for institutions seeking to implement data-driven, individualized academic planning and intervention strategies.