<p dir="ltr">This project focuses on detecting anomalies in satellite telemetry data using deep learning techniques, specifically autoencoders. Anomalies in telemetry can indicate unusual satellite behavior or system faults, which are important to detect early for mission safety and maintenance. </p><p><br></p><p dir="ltr">A synthetic telemetry dataset with 10 sensor variables and 8000 samples was generated, including 3% injected anomalies. After preprocessing, an autoencoder was trained on normal data and evaluated on mixed data. The model achieved a ROC-AUC of 1.000, Precision of 0.580, Recall of 1.000, and F1 score of 0.734. ROC curves, feature contribution charts, and heatmaps were produced for interpretability. </p><p><br></p><p dir="ltr">This project includes code, dataset, figures, and a research report. It demonstrates how simple deep learning techniques can be applied in space technology to detect faults. All resources are open-source and reproducible.</p><p><br></p>
