PSO-Learned Artificial Neural Networks for Activity Recognition

Abstract

The purpose of Activity Recognition (AR) is to recognize human activity using a sensor to get the data needed. Then, a machine learning approach is used to determine the type of activity performed. A machine learning technique often used in the classification problem is Artificial Neural Network (ANN), which is trained using a backpropagation algorithm. Although this technique has been significantly developed, it still has a few disadvantages compared to others. One of the disadvantages of the ANN is that the result is not always optimum because of randomized initialization and epoch limit. In this paper, a Particle Swarm Optimization (PSO) is proposed to train the ANN. Some experiments on a dataset of 10 k activities with six imbalanced classes show that the PSO-based ANN produces effectiveness of 100% and an F1 score micro of 0.88, which are much higher than the back propagation-based ANN that gives the effectiveness of 75% and an F1 score micro of 0.87.