DEEP RNN BASED HUMAN ACTIVITY RECOGNITION USING LSTM ARCHITECTURE ON SMARTPHONE SENSOR DATA

Abstract

Artificial Intelligence is gaining its prominence in areas like health, science, security etc. Deep Learning which is one of the most significant part of AI has been the focal point to analyze the patterns of different data that is related to these fields. It is very important to compute and assess the right data because enormous amount of data is present and also is generated which could be repetitive and also duplicate. This paper attempts to give an insight about how smart phone sensor data can be effectively analyzed through deep learning by using Long-Short-Term-Memory (LSTM) which is a building unit of Recurrent Neural Network (RNN). As Smart phones have become an important asset in our day-to-day lives, it could be an important catalyst to analyze human actions at various instances of time. For that purpose, Human Activity Recognition (HAR) dataset which is assessed on real-time human actions is considered and processed by using LSTM architecture. Overall, on Human Activity Recognition dataset, LSTM had approximately 4% more accuracy when compared to that of CNN for the same number of epoch values. Such sort of works are generally applicable to monitor people suffering from severe injuries, to estimate the health of the individuals, to provide some basic information regarding health before consulting a doctor. Human Activity Recognition, Deep Learning, Long-Short-Term-Memory, sensor data, Recurrent Neural Network.