Human Activity Recognition based on Semi-supervised Learning
| Autor: | CHEN, YAN-HAO, 陳彥豪 |
|---|---|
| Rok vydání: | 2018 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 106 Supervised learning is an active area of research that introduces a lot of computational problems. Therefore, there are many activity recognition systems that use machine learning tools to build training models to analyze and predict them through generated identification models. In the field of machine learning, analysis looks at the patterns through a collection of examples, and the collection of these instances is called a training data set. In the study of activity recognition, each instance represents a feature vector obtained in the sensing signal over a period of time. In the training data set, each data may be an action that has been classified or unclassified. For example, the data is calculated to know that the category is walking or running. However, in some cases, it is not possible to have the data automatically categorize the instances, because the data needs to be manually checked by the researchers, and the categories to which the instances belong are assigned according to their experience. Such a process is used in data mining applications. It is usually boring, costly and time-consuming. In other systems, some data is also unclassified. They are called semi-supervised learning systems. This research provides a new semi-supervised learning method to improve the identification performance. The experimental results have successfully improved the recognition rate of many classifiers. In order to study semi-supervised learning, we study various semi-supervised learning techniques proposed in recent years, and understand their operational processes and experiment results. Finally, we propose the Self-Adaptive Proportional Inclusion Assumption (SAPIA), modified in semi-supervised learning and activity recognition processes, through public data set experiments, and observed the classification performance under different classifiers in the past semi-supervised learning methods. Through the observation of the experimental results, we found that SAPIA's performance in a small number of training data sets will have better effects than other semi-supervised learning methods. Finally, based on the experimental results, our method will provide improvement in the field of semi-supervised learning and activity recognition. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|