Predicting Library Book Demand Using Ensemble Learning: A Comparative Study of Random Forest, Gradient Boosting, and Decision Tree Regressors

Accurate forecasting of library book demand is essential for efficient collection management, budgeting, and ensuring reader satisfaction. This study presents a comparative analysis of three ensemble learning algorithms—Random Forest, Gradient Boosting, and Decision Tree—for predicting the number of times a book will be borrowed in the next month. Using a synthetically generated dataset of 1000 library records enriched with engineered features (e.g., book age, availability ratio, borrowing rate, seasonal indicators), we evaluate model performance using R², RMSE, and MAE. Hyperparameter tuning was performed on the Random Forest model via grid search with cross-validation. Results indicate that Linear Regression achieved the highest test R² (0.5977), while the tuned Random Forest yielded an R² of 0.5430, demonstrating moderate predictive capability. Feature importance analysis revealed that popularity score, previous borrowings, and availability ratio are the most influential predictors. The study provides a practical framework for implementing data‑driven demand forecasting in library settings and includes a saved model artifact for deployment.