HIV-Infected Patients\' Daily Activities Predictions Using Smartphone Sensors

• Main Authors: Patrick Purnama, 王偉亮
• Other Authors: Ray F. Lin
• Format: Others
• Language: en_US
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/una3sn

碩士 === 元智大學 === 工業工程與管理學系 === 107 === HIV (Human Immunodeficiency Virus) or AIDS (Acquired Immunodeficiency Syndrome) is one of the world’s dangerous disease and it can be transmitted easily through sexual intercourse. Consuming ART (Antiretroviral) medicine right before HIV-infected patients want to do their sexual activities is one of the solution to prevent the HIV-transmission. However, there are difficulties to apply ART to HIV-infected patients regularly. One of those difficulties is the patients always forget to consume their ART medicine right before they want to do their sexual event. Hence, this research aimed to determine HIV-infected patients’ sexual activities by predicting their travel behavior, APP usages and the preparation point of their sexual activities based on their smartphones’ sensor data. This smartphone’s sensor data includes mean and standard deviation of their smartphone sensors (including accelerometer, gyroscope, and orientation sensor), distance, and speed data. Those smartphone sensors’ data were collected using self-developed APPs for smartphones under Android System. For predicting HIV-infected patients’ travel behavior and APP usages, this research used one of machine learning algorithm, that algorithm was Random Forest. Meanwhile, for determining the preparation point of HIV-infected patients’ sexual activities, this research built selfdeveloped algorithms, which divided into two types: algorithm for determining HIV-infected patients’ sexual activities outside their house and algorithm for determining HIV-infected patients’ sexual activities inside their house. The result for predicting HIV-infected patients’ travel behavior showed that the highest accuracy was obtained by using accelerometer sensor only with 92.20% accuracy and the activity with the highest accuracy was Riding Scooter. The result for predicting HIV-infected patients’ APP usages showed that the highest accuracy was obtained from using orientation sensor data only and using APP categorization based on their data type with 76.01% accuracy. Meanwhile, most of the APP categories were confused with Social Media APP because their data type was similar with Social Media APP’s data type. The last one, for determining HIV-infected patients’ sexual activity, this research could determine 6 out of 8 sexual activities outside their house which already determined by theiv researcher before based on GPS location. However, the algorithm for predicting sexual activities inside patient’s house could determine 22 possible sexual activities which already defined by the researcher before according to sensor data analysis.