Classification of Human Activities using data captured through a smartphone using deep learning techniques

Classification and recognition of activities performed by humans is an essential task when we are into an era of connected-sensors and Industry 4.0, also referred to as the Internet of Things (IoT). In this paper, we propose an intelligent human activity-recognition system that can automatically predict daily human activities based on data acquired by sensors, using deep learning techniques like Two-dimensional Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM). The data collected using the sensors is fed to CNN and LSTM network. The proposed model is trained and tested on the WISDM dataset. The proposed deep learning architecture has Batch Normalization (BN) and dropout layers to avoid overfitting. The proposed architecture is of low complexity and achieves high accuracy and F1 score. The final accuracy obtained using the LSTM model is up to 97.5% with 21,922 trainable parameters. The obtained results are represented with the help of a convolutional matrix and learning curves. The results are quite promising and can be utilized without any restrictions on environments or domain-structures.