Evaluation of Principal Component Analysis Algorithm for Locomotion Activities Detection in a Tiny Machine Learning Device

Abstract

Data generated by human locomotion activities is a process that involves the analysis of hundreds or thousands of data in a reduced time, due to the very nature of the signals generated and the techniques for implementing classification models and event detection. It is advisable to try to reduce the number of characteristics or select the most important elements of the captured signals showing, in this paper, the use of principal component analysis (PCA) techniques. The use of machine learning techniques for reduced hardware devices in intelligent environments, allows generating a solution for the non-invasive supervision of activities, complementing the use of PCA with other classification algorithms suitable for the treatment of data with a high number of characteristics such as support vector machines (SVM). Therefore, the evaluation of PCA processes and SVM algorithms is shown, selecting the one that has the best performance during its implementation in IoT devices with low hardware resources. Finally, it is considered that the memory space consumed in the IoT device and the execution time of the processes are critical elements to make the comparison and contrast of the PCA models, allowing to select and develop a reliable and efficient model in small devices of IoT.