AlertSafeNet a multichannel deep learning framework for real time anomaly detection using wearable device data

This research investigates the integration of wearable device data analysis to enable proactive safety monitoring within intelligent systems. The objective is to develop a deep learning framework capable of automatically detecting anomalous gait patterns from the wearable device data and generating alerts in real-time. Smartwatches and fitness trackers, through their built-in sensors, continuously record physiological and contextual signals including heart rate, SpO₂, ECG, location, and more. Analyzing this multichannel time-series data lets the proposed framework flag abnormal behavioral patterns that could signal a safety risk. The proposed framework enables the identification of abnormal behavioral patterns that may reflect potential safety risk scenarios. In high-risk scenarios, such systems can activate real-time alerts to trusted contacts or concerned authorities. In this work, we propose AlertSafeNet, a multichannel 1D convolutional neural network with residual connections, designed to learn temporal dependencies and detect anomalies from wearable sensor data. The dataset is created using smart watch-derived physiological signals combined with contextual information from smartphones. Once trained, the model runs in a cloud-based environment, supporting real-time inference and continuous monitoring. The results demonstrate promising performance: the model reaches 97.9% classification accuracy, with strong precision, recall, and F1-score — a good indicator of its usefulness for real-time anomaly detection in safety-critical settings.