This document presents a machine learning-based system for detecting ATM fraud. It examines current types of ATM fraud and proposes using machine learning algorithms to build a detection system. The study evaluates supervised, unsupervised and semi-supervised machine learning approaches for this task. It specifically explores hidden Markov models, support vector machines and decision trees for fraud detection. The document outlines the methodology, including data collection, model training and testing. Charts and figures show sample outputs from the system, including detected fraud over time. The system is found to effectively detect fraud while minimizing false positives. Machine learning is concluded to be well-suited for ATM fraud detection.