WISNet: A Deep Neural Network based Human Activity Recognition System

Nowadays, human activity recognition (HAR) is a key component of many ubiquitous innovative solutions, where both accelerometer and gyroscope data provide information about an observed person's physical activity. HAR offers a diverse variety of important applications, including healthcare, burglary detection, workplace monitoring, and emergency detection. In this paper, we propose a custom 1D-CNN deep learning approach named WISNet to recognize complex human activity. The proposed model is compared with other time-series deep learning models, which include Gated Recurrent Units (GRU), Long Short-Term Memory (LSTM), and recurrent neural network (SimpleRNN), to classify and detect the six class WISDM dataset. This dataset includes basic ambulation-related activities, like walking, jogging, and climbing stairs; having been retrieved from sensors in smartphones and smartwatches. The proposed WISNet architecture outperforms the other time series models in detecting the six classes; GRU, LSTM, and SimpleRNN attain an accuracy of 95.27%, 95.15%, and 91.8%, respectively. For HAR, the proposed trained WISNet model achieves enhanced an accuracy and F1-score of 96.41% and 0.95, respectively, surpassing the other models. The proposed WISNet architecture consists of a smaller number of convolutions, generating almost less than 10% features when compared to the GRU, LSTM, and SimpleRNN, presenting also better results.