Machine Learning Applied To Fall Detection in the Elderly
Resumo
Context: With the increase in falls among the elderly, rapid detection becomes essential to reduce fatal risks. This challenge stimulates advancements in machine learning algorithms and IoT technologies in medicine, aiming to improve the safety and longevity of the elderly. Problem: An increase in falls among the elderly is a domestic severe health issue. Solution: Development of an efficient machine learning algorithm using accelerometer data from wearable devices to detect falls in the elderly. SI Theory: The study employs General Systems Theory, not to develop new hardware devices but to integrate and analyze data from accelerometers that already exist in wearable devices. It focuses on the synergy between this hardware data and software algorithms and their interaction with human behavior to enhance the detection of falls in the elderly. Method: This work uses the applied methodology to evaluate KNN, Decision Tree, and MLP algorithms applied to accelerometer data, focusing on accuracy and efficacy. Summary of Results: The MLP model stood out with high efficacy in fall detection, achieving a recall of 97.92% during the testing and 100% during the validation phases. This indicates the model’s strong ability to identify falls correctly, a crucial factor for the safety of the elderly. Contributions and Impact in the IS area: Presents an efficient solution for the health of the elderly, with the potential to reduce accidents and improve quality of life. Highlights the importance of validation in natural environments and with diverse individuals for future research.
