umbertogriffo/Predictive-Maintenance-using-LSTM

Implements both regression and binary classification approaches on NASA turbofan engine sensor data, processing 21-channel multivariate sequences to predict remaining useful life or failure within time windows. Built with Keras/TensorFlow and includes data preprocessing pipelines for temporal sequences across multiple engine time series, achieving 0.97 accuracy on failure classification and MAE of 12 cycles on regression tasks. Demonstrates extensibility to multi-class time-window problems through customizable data labeling strategies.

722 stars. No commits in the last 6 months.