ML-XRay
Created a Failure Atlas that maps each algorithm to the conditions that break it and why.
From-Scratch ML Algorithms, Stress-Tested to Find Their Failure Conditions
Project Overview
ML-XRay builds machine-learning fundamentals at the mechanism level. Core algorithms are implemented from scratch, then deliberately stressed with controlled data changes to reveal exactly where and why they fail.
Project Description
Each experiment follows a predict, break it, explain why loop: a hypothesis is written before the run, then exactly one data property is changed. The project documents ten experiments, capped with a seven-algorithm side-by-side stress test and a concise Failure Atlas.
- Built Linear Regression, Logistic Regression, KNN, Decision Tree, K-Means, and PCA from scratch.
- Studied Random Forest, SVM, and Gradient Boosting experimentally alongside the from-scratch implementations.
- Changed one factor at a time: noise, class overlap, irrelevant features, unscaled data, non-convex clusters, label noise, class imbalance, and small samples.
- Condensed the findings into a Failure Atlas: algorithm, failure condition, and the mechanism behind it.
Technologies Used
PythonNumPyMatplotlibMachine Learning
Project Links
Report Slideshow
