<p>Noise robustness under IoT sensor perturbation.</p>

Common approaches employ wearable or ambient sensors, signal preprocessing, feature extraction, and deep learning models such as CNNs, LSTMs, BiLSTM, and attention-based classifiers. These techniques generally achieve high classification accuracy, robust temporal modeling, and improved detection of complex activities, supporting reliable automation, health assessment, and anomaly recognition.

Challenges include sensor noise, data imbalance, privacy concerns, limited generalization across homes, computational complexity, and reduced performance in real-time or resource-constrained environments. The proposed IoT framework integrates normalized sensor vectors with BiLSTM and attention mechanisms, achieving improved temporal feature learning and more accurate activity and anomaly classification. The proposed BiLSTM-Attention framework achieved 97.2% accuracy, 95.1% Dice coefficient, 90.6% Jaccard index, 96.5% sensitivity, and 98.1% specificity on the smart home activity dataset.</p></div>