Gait Activity Authentication Using LSTM Neural Networks with Smartphone Sensors

Abstract

With increasing usage of smartphones in people's daily life, the demand for secured access control becomes more and more urgent since people tend to store their personal information, such as personal identifiers and bank account details on their smartphones. In this work, we present a novel framework for user authentication techniques based on human gait related activities analyzed from smartphone sensor data. We propose using LSTM Neural Network which is suitable to automate feature extraction from raw sensor inputs for human activity recognition. Furthermore, we investigate the impact of activity dependent authentication which produces better accuracy than activity independent authentication. The results show that the proposed model is promising compared to other traditional machine learning classifiers and outperforming some of the previous reported results.