<![CDATA[The increasing elderly population in Indonesia presents a significant challenge in ensuring safety, particularly concerning the risk of falls, which can lead to serious injuries. This study aims to design and develop a fall detection device based on the Internet of Things and a Feedforward Neural Network algorithm. The system was built using an ESP32 microcontroller and ADXL345 accelerometer sensor, and integrated with firebase and nodered to enable real time fall notifications via whatsapp. The SisFall dataset was utilized, consisting of two primary classification labels fall and activities of daily living . Data processing and model training were carried out using the Edge Impulse platform. Three different models were evaluated, and Model 1 was selected as the best-performing model after hyperparameter tuning, achieving the highest accuracy of 99.7% and the lowest loss of 0.02. The model was quantized to int8 format for efficient edge deployment. Real world testing of the device involved 30 movement variations, each tested five times. Results showed that the system achieved a classification accuracy of 95.76%, precision of 97.22%, sensitivity of 93.33%, and specificity of 97.78%. These metrics indicate that the fall detection system was successfully developed and demonstrates excellent and reliable performance for practical applications.]]>
Copyrights © 2025
