Student Performance Prediction for Adaptive Learning using Improved COOT Optimization Algorithm

Predicting student performance is crucial for both preventing failure and enabling personalized teaching-and-learning strategies. However, the presence of irrelevant features, selecting the features is difficult which was led to overfitting and unreliable predictions. It affects the model to focus on unnecessary patterns rather than key dropout indicators. To solve these problems, this research proposes improved COOT optimization algorithm used for selecting most relevant features for both temporal and contextual analysis thereby improving the overall prediction accuracy for adaptive learning. Initially, student data obtained from a student performance dataset then, Min-max normalization in preprocessing is to align feature distributions across different academic years by standardizing mean and variance, mitigating distribution shifts in long-term data. Feature selection through the COOT algorithm effectively addresses the high dimensionality of student learning data by identifying key features that contribute to the evaluation of student assessment performance. The proposed COOT algorithm achieves an accuracy of 98.32% when compared with existing methods namely stacked generalization for failure prediction to demonstrating its evaluation of the effectiveness of adaptive learning for college students.