Details

This project provides a from-scratch implementation of a Gradient Boosting Classifier for binary classification, inspired by Sections 10.9–10.10 of The Elements of Statistical Learning. Built without external libraries, the model iteratively refines predictions by training shallow decision trees on the negative gradient of the logistic loss. The project includes synthetic and real-world dataset evaluations (IBM Attrition), visualization of decision boundaries, ROC curves, learning curves, and confusion matrices, along with adjustable hyperparameters for tuning model performance.