Exploring Elderly Activity Recognition Using Deep Learning Through Sensor-Enabled Android App Data Collection

Recognition, tracking, and classification of human activity is a crucial development in assisted living technologies that can support older individuals with their daily activities. Techniques based on vision or sensors are used in conventional approaches to activity recognition. Activity recognition is helpful in tracking elderly people. In this work, we present a novel cultural and regional Elderly Activity Recognition system. We collected the dataset through a Sensor-Enabled Android App that records various characteristics of the movement, using a sensor including the accelerometer, gyroscope, and linear acceleration. We applied Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN), and a Hybrid model in our research to recognize elderly activity. The hybrid model appeared as the best-performing classifier, achieving a reliable level of precision, recall, and Fl score, with outstanding accuracy of 91.09 %, and exceptional performance indicates the Hybrid's better capacity for accurately Elderly Activity Recognition system using Sensor-Enabled Android App data. Nevertheless, the accuracy ratings of 88.99% and 90.96% acquired by LSTM and CNN, respectively, also exhibited good categorization. LSTM and CNN models had significantly lower accuracy scores than the Hybrid model, but these determinations show how acceptable the classifiers are at spotting elderly activity recognition. The research's other novelty is making an Android app for data collection, which is very flexible for future researchers as they can add and remove new activity for collecting the dataset.