Predictive Maintenance
A binary classifier that reads live sensor data — temperature, rotational speed, torque, and tool wear — and predicts whether an industrial machine is about to fail, so maintenance can happen before a breakdown does.
Trained on the AI4I 2020 dataset (10,000 machine readings, roughly 3.4% true failures), the project's real challenge wasn't accuracy — it was building a model that still catches rare failures despite heavy class imbalance, and can explain every prediction it makes.
PythonPandasscikit-learnRandom ForestLogistic Regressioneli5joblibFastAPI
Model accuracy, held-out test set
Random Forest98.64%
Logistic Regression97.52%
10K
Sensor readings
3.4%
True failure rate
6
Input features
2
Models compared