Dominance of deep LSTM in smartphone sensor based human activity classification

Abstract

Human activity categorization is one of the challenging topics on which several types of researches have been conducted. With the improvement of modern technologies and algorithms, there have been many plausible solutions. However, earlier studies had been limited to a few activities based on indoor situations. Conducting the classification of complex continuous activities was an interest off time. As the world progresses, it needs more accurate data visualization in real-time for medical service purposes. Even this prediction can become crucial for the business aspects. For this purpose, in this research, the number of activities have been increased for classification and is attempted to detect with LSTM, XGBoost, SVM and Neural Network techniques. In this work, a smartphone app was developed that performs activity recognition using the embedded accelerometer, gyroscope sensors and based on the selected features, classify activities, especially with distinct signatures.