An Ontology of Machine Learning Algorithms for Human Activity Data Processing

Machine learning algorithms are the main tools in the field of data analysis. However, extracting knowledge from data sets originating in real life requires complex data processing. Obtaining the available tidy data sets and selecting the appropriate analysis algorithm are important issues for data analysts. Because of the complexity of the dataset and the diversity of the algorithms the researchers take too much time in selecting and comparing these algorithms. Human Activity Recognition is a typical example in Internet of Things. Its principle is to identify human behavior by analyzing the coordinate data from the sensors on the human body so that we can achieve remote monitoring. A precise Human Activity Recognition application can serve as a real-time monitoring of the elderly or vulnerable behavior. However, due to the unpredictability of human behavior, these sensor data require relatively complex processing. Therefore, we propose an ontology-based algorithm recommendation system. It consists of several parts: algorithm pool, data features, model features, and mathematical theory. The framework provides data researchers with reasonable solutions based on the characteristics of the data set and the task requirements. Especially for the Internet of Things data such as Human Activity Recognition data set, its recommendations can save users much time for analysis and comparison