Using Smartphone-application-usage Features for Predicting the HIV Patient’s Daily Emotional States

• Main Authors: Shang-Ru Yu, 余尚儒
• Other Authors: Ray F. Lin
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/p3x6xs

碩士 === 元智大學 === 工業工程與管理學系 === 107 === In recent years, there is an increasing number of HIV-infected people. In order to improve the quality of care given by patient managers, the emotional values of HIV patients will be detected through regression models using the characteristic values of mobile phone software. According to statistics from Taoyuan Hospital, the average number of people attending the Infectious Diseases Department in Taoyuan Hospital is about 2,400 annually; however, there are only 6 in the number of patient managers. There is obviously insufficient manpower to provide high quality of care. Furthermore, the average HIV-infected person only visits the doctor once every three months. Whenever there is no active contact during the period, it is difficult to detect when the patient is encountering difficulties and is need of care. Due to disease-related stress, HIV-infected people are prone to induce depression, and depression is not easy to be detected. Therefore, the main purpose of this study is to use the characteristics of mobile phone software. The value detects the emotional state of HIV patients. In past studies, Ballve (2013) and Wiese et al. (2015) pointed out that the use of mobile phones is a traceable behavioral feature that may be related to emotional state. Also, van Breda et al. (2016) with Shapsough et al. (2016) used machine learning methods to achieve good prediction results. Therefore, this study will find suitable eigenvalues and prediction methods through two-stage analysis. The first stage analysis only selected the data of one subject and calculates the relevant eigenvalues of the time of using the mobile phone software. After using four-machine learning methods, the mean square error evaluation model is used to obtain the best method as the decision tree model. In this stage, analysis is done through selection four subjects and according to the data collection status. Afterwards, analyze the individual subject data and the overall subject data and use the mean square error and R-square value to judge the model performance. In the analysis of individual subjects before and after the mobile phone software grouping, the decision tree was used to compare with the stepwise regression analysis method. In the whole subject analysis, the mobile phone software was grouped and then standardized before and after data. The decision tree and stepwise regression analysis method were also compared. The results of the study found that in the second-stage analysis, the model of the mobile phone software group was better in the analysis of the data of the individual subjects. The mean variance of the two subjects was 0.0881 and 0.392 respectively; the R-SQ values were 99.3% and 59.6 respectively. The analysis of the overall subject data includes information on four subjects. The model before the standardization of mobile phone data performed well, with a mean square error of 0.4231 and an R-SQ of 40.6%. Obtaining the important eigenvalues of the model from the stepwise regression analysis, we can understand that the use of mobile phone software affected by the mood fluctuation of each subject is not the same. Therefore, this study suggests that the subject’s mood can affect the type of APPs that he uses. The categorization of subjects according to their usage of different mobile phone APP types is expected to improve the overall model performance of the subjects in the future.