Back to Projects
Sensor Data Anomaly Detection
Machine LearningData SciencePython
Role
Data Scientist
Tech Stack
PythonScikit-LearnPandasMatplotlib
The Problem
Detecting rare equipment failures in noisy sensor data is critical to prevent downtime but difficult due to class imbalance.
The Solution
Engineered a hybrid ML approach using Isolation Forests and Autoencoders. Preprocessed high-dimensional time-series data to isolate signal noise.
Impact & Metrics
Achieved 92% F1-score on test dataset, significantly reducing false positives compared to baseline thresholding.