Elderly people are living alone most of the time during daytime without manual monitoring, and they are very much prone to sudden fall. HARHuman Activity Recognition (HAR) plays a pivotal role in recognizing human activities through seamless, continuous, and noninvasive monitoring of ADLs to provide critical insights into an individual’s physical and cognitive health. Activity recognition using single sensor data is questionable in terms of reliability due to disturbances through different factors such as noisy signal, environmental hazards, position of sensor, and sensor calibration. Hence, combining multiple heterogeneous sensor data, i.e., using sensor fusionSensor fusion method, effects of these external hazards on HARHuman Activity Recognition (HAR) model performance could be reduced, thus accuracy will increase, and context-aware HARHuman Activity Recognition (HAR) could efficiently be implemented. Sensor fusionSensor fusion-based HARHuman Activity Recognition (HAR) models can evaluate human movement patterns and physiological states of health condition simultaneously and can find relation with the health condition and deviation in the movement pattern. Healthcare providers hence get the insight to identify behavioral pattern shift that may be used as marker of early symptoms of chronic diseases, e.g., neurodegenerative disorders such as Parkinson’s disease, Alzheimer’s, gait issues, musculoskeletal disorders, and cardiovascular disease conditions. Fusion methods could be at any level: data level, feature level, information level, etc., utilizing deep learningDeep Learning (DL) models such as CNNConvolutional Neural Networks (CNN), RNN, GNN, and LSTMLong Short-Term Memory (LSTM) for disease predictionDisease prediction by capturing long-term activity patterns and contextual relationship.