Integrating Sensors Data and Machine Learning for Enhanced ADL and Fall Detection Systems

Developing a fall-proof environment presents various challenges. Statistics show that falls are the leading cause of injury-related death among people aged more than 80. Fall detection and prediction systems can increase stability and safety to reduce the risk. One of the best ways to develop fall detection/ prediction systems (FDPS) is with multi-sensor (accelerometer, gyroscope, magnetometer) data by using machine learning algorithms. However, the challenges for getting this system are a selection of optimal sensors, a fusion of sensors and machine learning (ML) algorithms for the Fall Detection and Prediction System. This study analyses the impact of sensor integration, ML algorithms performance on integrated sensors, and individual data for the FDPS. To achieve the objective of this research, five publicly available multi-sensor datasets are used; a proposed feature selection algorithm is applied for sensor data selection and ML algorithms such as Decision Tree, Naive Bayes, K-Nearest Neighbour, Support Vector Machine, Adaboost, Gradient Boosting, and Random Forest are applied for fall prediction. The ML algorithm’s performance is significantly enhanced by using integrated sensors, achieving an accuracy of 100% on dataset d3. Among the algorithms tested, the Random Forest algorithm emerged as the best-performing model across all five datasets, d1 to d5, with an accuracy of 85%, 98%, 100%, 97%, and 41%, respectively. Additionally, evaluating the performance of individual sensors for predictions, the accelerometer was found to be the most effective sensor.