FinHealth: Passive assessment of mental health using financial (and other) data via machine learning
According to the World Health Organization (WHO), 1 in 8 (WHO, 2022) and about 970 million people are living with mental disorders (WHO, 2019). About two-thirds of mental illnesses go untreated, and it takes years before the mental illness is detected. Furthermore, the WHO reported that after the COVID-19 pandemic, mental disorders such as anxiety and depressive disorders increased by 25% (WHO, 2022) and many of these people do not have access to treatment. This project aims to assess financial and other data, to ascertain how relevant this data is to individual mental health. In addition, machine learning models will be built to detect the onset of mental illness using this data. Furthermore, this will be done using a privacy-sensitive methodology, to build the trust of the patient or users. The project will focus primarily on bipolar disorder and student mental health. People who have bipolar disorder, a very serious mental illness affecting about 1% of people in the world, are particularly affected
by challenges with finances during episodes (Merikangas et al., 2011). They experience extreme variations in how they feel, ranging from a very low depressed mood to the ecstatic, top of the world, happy highs associated with manic states. It is during these high periods that people with bipolar disorder exhibit impulsive behaviour and often make large impulsive purchases. Spending sprees have long been associated with bipolar disorder.
Prior work in the field in both bipolar disorder and student mental health has centered on cross-sectional self-reported data. This project will focus on mining personal financial data in a privacy-sensitive manner in order to predict mood and mental health states (as measured by self-reported and clinician rated scales). In this project, a systematic literature review will be conducted to ascertain the issues and best practices for securing and authenticating personal financial data, in a way that preserves the privacy of such sensitive data without losing the mental health signal. Furthermore, the project aims to identify non-identifying features that track mental health status (e.g. meta-data like time of spending, categories
of spending, normalized spending patterns). Thereafter, machine learning models will be implemented to predict moods and mental health state from personal financial transactions.