AI – Based Data Analytics in Bluetooth Smart Sensor Networks

Abstract

The growing adoption of Internet of Things (IoT) devices demands intelligent and energy-efficient data analytics in Bluetooth Low Energy (BLE) smart sensor networks. Traditional cloud-based approaches suffer from high communication overhead, privacy concerns, and latency limitations unsuitable for real-time monitoring. This paper proposes an AI-driven framework for anomaly detection in BLE smart sensor networks using a hybrid CNN-LSTM autoencoder combined with a Federated Learning protocol organized across a three-tier IoT architecture — BLE sensor nodes (Nordic nRF52840), an edge gateway (Raspberry Pi 4), and a cloud analytics server. The CNN extracts spatial features from sensor signals while the LSTM captures temporal dependencies in time-series data. The Federated Averaging (FedAvg) algorithm enables local model training without transmitting raw data, preserving privacy and reducing communication cost. Evaluation on the SKAB, MIMIC-III, and IBRL benchmark datasets achieved an F1 score of 0.964, a 38% reduction in communication overhead, 47 ms average inference latency, and extended battery life from 12 to 41 days. These results confirm that the proposed framework is an effective, privacy-preserving, and resource-efficient solution for real-time anomaly detection in IoT-based BLE smart sensor networks.