With the increasing aging population, the significance of daily living ability training systems for the elderly has become increasingly important in their daily lives.The overall goal of this project is to learn different data mining methods and find ways to improve the detection accuracy of human daily behavior, especially for falling behavior. Specifically, this article mainly analyzes the results of three existing data mining methods applied to specified datasets, and greatly improves the detection accuracy of human daily behavior through new experimental methods. Experimental results have shown that the improved learning algorithm outperforms existing machine learning algorithms to a large extent. At the same time, this article will organize this into a set of training and application methods, which not only includes the training data and classifier acquisition part, but also includes the algorithm part of applying the classifier, making the new method universal.