Dataset and Python Scripts for Student Dropout Prediction Using Explainable Machine Learning and LMS Analytics

This repository contains the supplementary materials associated with the article “Comparative Evaluation of Explainable Machine Learning Models for Student Dropout Prediction Using LMS Analytics.” The materials include the original dataset, the preprocessed dataset, and the Python scripts used for data preprocessing, model training, performance evaluation, and explainability analysis using SHAP. The purpose of this repository is to support transparency, reproducibility, and reuse of the computational workflow implemented in the study. The research focuses on predicting student dropout in virtual learning environments using Learning Management System (LMS) interaction data, academic variables, and Explainable Artificial Intelligence techniques. The dataset includes academic, sociodemographic, and LMS interaction variables such as course, age, gender, socioeconomic level, employment status, semester, login frequency, forum participation, task submissions, late submissions, connection time, resource views, final grade, and dropout status. The target variable indicates whether a student remained active or dropped out. The Python scripts were developed to implement the complete Machine Learning workflow, including data cleaning, categorical encoding, feature scaling, outlier analysis, model training, evaluation metrics, ROC curve generation, confusion matrix analysis, SHAP global feature importance, SHAP summary plots, SHAP dependence plots, individual explanation plots, and numerical importance tables. Repository Contents Processing.py This script performs the data preprocessing workflow. It loads the original dataset, manages missing values, supports categorical and numerical imputation, applies outlier detection methods, encodes categorical variables, scales selected numerical variables, removes missing rows when necessary, and exports the final processed dataset. Main functions include: Loading the original dataset. Creating a working copy of the data. Handling missing categorical and numerical values. Applying Label Encoding to categorical variables. Applying StandardScaler to numerical variables. Detecting and removing outliers using IQR or Isolation Forest. Exporting the cleaned dataset as datos_preprocesados.txt. Script for Model Training.py This script implements the Machine Learning model training and evaluation process. It loads the preprocessed dataset, separates predictor variables from the target variable, performs a stratified train/test split, trains the classification models, computes performance metrics, generates visual evaluation outputs, and saves trained models. The evaluated models include: Logistic Regression K-Nearest Neighbors Decision Tree Random Forest XGBoost The evaluation metrics include: Accuracy Precision Recall F1-score ROC-AUC Cross-validation score Confusion matrix Classification report SHAP Explainability.py This script performs the explainability analysis using SHAP. It loads the processed dataset and the trained XGBoost model, constructs a SHAP TreeExplainer, calculates SHAP values, and generates visual and numerical explanations of the model predictions. The SHAP analysis includes: Global feature importance plot. SHAP summary plot. SHAP dependence plots for predictor variables. Individual waterfall explanation. Numerical SHAP importance table. This script helps identify the variables that most strongly influence dropout risk prediction, including final grade, resource views, task submissions, forum participation, connection time, and other LMS interaction indicators. dataset_abandono.txt This file contains the original dataset used in the study. It includes raw academic, sociodemographic, and LMS interaction records before preprocessing. The target variable indicates student dropout status. The dataset includes variables such as: Student identifier Course Age Gender Socioeconomic level Employment status Semester Login frequency Forum participation Task submissions Late submissions Connection time Resource views Final grade Dropout status datos_preprocesados.txt This file contains the processed dataset used for Machine Learning model training, validation, and SHAP-based interpretability analysis. The dataset includes cleaned, encoded, and transformed variables suitable for classification algorithms. The processed file removes the student identifier and represents the modeling variables in numerical form.