Learning Disability Diagnosis Using Advanced Preprocessing and Ensemble Classification Techniques

Abstract

Early detection of learning disabilities (LD) is essential for prompt intervention, yet it is still difficult because LD signs are subtle and variable. Previous research frequently ignores the significance of class imbalance and data preprocessing, leading to skewed or inadequate forecasts. This paper suggests a hybrid machine learning framework for LD diagnosis that combines several classification algorithms with thorough preprocessing, such as feature scaling, SMOTE-based class balance, missing value management, outlier removal, and feature selection. The effectiveness of a stacking ensemble that incorporates the advantages of Random Forest, XGBoost, LightGBM, and a Multilayer Perceptron is assessed. The efficiency of combining improved preprocessing with ensemble learning is demonstrated by the experimental findings, which show that the stacking ensemble greatly outperforms individual classifiers in terms of accuracy, recall, F1-score, and ROC-AUC. The framework facilitates prompt educational support and intervention by offering a strong and dependable instrument for early LD detection.