Early Detection of Emotional Issues in High School Students Through Statistical Analysis of Academic Data

The goal of this work is to identify markers for the onset of emotional issues, such as depression and anxiety, in high-school-age children. Such markers may involve declining academic performance, truancy, and behavioural issues in schools. The ability to identify markers predicting the onset of emotional issues in school-age children would allow schoolteachers and administrators to notify parents and guardians of the children affected as soon as the markers are detected. This would in turn facilitate early diagnosis and treatment for children who are indeed developing emotional issues. To accomplish our goal, we applied unsupervised learning methods to partition the input dataset into clusters with similar characteristics with respect to potential markers, such as declining or fluctuating academic performance. Our analysis seeks to cluster students into three classes (``critical", ``monitor" and ``stable") whose size aligns with statistics available, for instance, from the US Center for Disease Control and Prevention (CDC). The student’s academic data are modelled as time series. Three main clustering approaches are evaluated empirically: (1) clustering of time series with different lengths, (2) classical clustering algorithms and (3) density-based clustering with feature reduction. As it turns out, only the last approach respects our desired cluster size while also achieving good accuracy results with respect to the markers. Students in clusters exhibiting fluctuations or sharp declines in academic performance are intended to be referred to their school counsellors for further investigation. We are currently working with a dataset of about 500 students from 5 Catholic high schools under the jurisdiction of the Diocese of Peoria in Peoria, Illinois.