Ship Engine Anomaly Detection

An anomaly detection project comparing statistical and machine learning methods to identify abnormal ship engine behaviour from sensor data.

Overview

This project explores unsupervised anomaly detection for predictive maintenance using ship engine sensor data. The analysis compares IQR, One-Class SVM, and Isolation Forest to detect unusual operating patterns.

Methods

• IQR
• One-Class SVM
• Isolation Forest
• PCA for visualisation

Key Result

Isolation Forest performed best overall as a practical anomaly detection method for this dataset, offering a strong balance of robustness and interpretability.

Files

• project_notebook.ipynb
• project_report.pdf

Tools

Python, pandas, NumPy, Matplotlib, Seaborn, scikit-learn

Summary

This project demonstrates anomaly detection, model comparison, and PCA-based visualisation in a predictive maintenance context.