Prediction and Explainability of School Dropout in Peru Using Machine Learning and Explainable Artificial Intelligence

This project introduces an advanced machine learning framework designed for predicting student dropout using AutoML and SHAP (SHapley Additive exPlanations). The primary objective is to predict students' likelihood of dropping out based on various demographic, academic, and behavioral features. The system integrates AutoML for automated model selection, hyperparameter tuning, and feature engineering, making it efficient and scalable for large datasets. A key part of the process involves using SHAP to interpret model predictions, ensuring transparency in identifying which features contribute most significantly to dropout risk. The dataset comprises various student attributes, such as attendance, grades, socioeconomic status, and engagement levels, which are preprocessed through standardization and encoding techniques. SHAP values are employed to provide an interpretable explanation for individual predictions, enabling educators and administrators to understand the factors driving dropout predictions. A comparative analysis of different AutoML models was conducted, including traditional machine learning algorithms like Random Forest, SVM, and ensemble models, with the aim of improving the predictive accuracy and generalization ability of the system. The proposed framework offers a powerful tool for educational institutions to proactively identify students at risk of dropping out, enabling early intervention and personalized support. This model can be adapted to various educational contexts and contributes to improving student retention rates, advancing the field of predictive analytics in education.

Keywords: Student Dropout Prediction, AutoML, SHAP, Predictive Analytics, Machine Learning, Feature Engineering, Model Interpretability, Educational Retention, Early Intervention, Dropout Risk, Ensemble Learning, Hyperparameter Tuning, Transparent AI, Data Preprocessing, Educational Analytics.