Performance Comparison of Machine Learning Algorithms for Human Activity Recognition

Human activity monitoring and recognition is one of the significant focuses for researchers recently due to the concerns regarding improvements of the healthcare, surveillance and many others that includes human-machine interactions. The data collected because of advances in sensor technology and the machine learning algorithms enable the work in the field of recognition of human activities to progress. Life quality of especially elderly people highly benefits from the activity monitoring and recognition research. This study aims to collect data from multiple activities and train multiple machine learning algorithms using the recorded data. For this purpose, a total of two wireless sensors were placed on the ankle and calf regions of the subjects on one leg, and the data obtained from these sensors were used in decision tree, artificial neural network, and deep learning algorithms. The performance of the algorithms was evaluated with the F1 score metric. The results show that the most suitable algorithm for this dataset is the deep learning algorithm with an average F1 score of 0.989.