Explainable IMU Anomaly Detection: Impact of Window Size on Machine Learning Performance under Noise Injection Açiklanabilir IMU Anomali Tespiti: Gürültü Enjeksiyonu Altinda Makine Ögrenmesi Performansina Pencere Boyutunun Etkisi
34th Signal Processing and Communications Applications Conference, SIU 2026, İstanbul, Turkey, 7 - 10 July 2026, (Full Text)
- Publication Type: Conference Paper / Full Text
- Doi Number: 10.1109/siu71813.2026.11636468
- City: İstanbul
- Country: Turkey
- Keywords: anomaly detection, explainable artificial intelligence, IMU sensors, machine learning, window size analysis
- Yıldız Technical University Affiliated: Yes
Abstract
In this study, anomaly detection is investigated under controlled noise injection using IMU sensor data collected from a mobile robot. The collected time-series data is processed using a windowing approach, and machine learning models are developed based on extracted features. Logistic regression and random forest algorithms are employed for anomaly detection, and SHAP is used to enhance model interpretability. In addition, the impact of different window sizes on model performance is analyzed. Experimental results indicate that window size has a significant impact on model performance.