Bipolar affective disorder (ChAD) is a serious mental disorder that affects more than 2% of the world's population. It is characterized by manic episodes of elevated mood and excessive activity, interspersed with periods of depression. It is very difficult to predict the next phase change of a ChAD patient's behavior. There is no way to alert patients and their physicians of an impending phase. The main goal of the project was to use a machine learning algorithm to create an innovative application that could collect and analyze information to detect and predict a patient's phase changes.
Phase change diagnosis can be facilitated by monitoring data collected by the patient's smartphone. The first version of the app we prepared collected data from features such as the number of incoming and outgoing phone calls, the length of SMS messages, the number of steps or voice features. Using the collected data, the application could use it to prepare a predictive model based on machine learning.
As part of the project's implementation, we researched various methods that would allow for definitive day-ahead prediction of changes in a patient's condition using supervised and unsupervised Machine Learning techniques.
The prepared application has an implemented version of the OpenSmile library, which collects data from the physical characteristics of the voice. Patients can also mark hours and assess their mood and monitor it in their daily life. The data collected from the patient's application is subjected to clustering using appropriate algorithms. This allows for a one-day prediction of phase change in a patient with bipolar disorder.
Technologies and tools
mongoDB, PostgreSQL, R