Detection of Behavioral Health Challenges in High School Students

The primary objective of this project is to develop a predictive model for the detection of behavioral health challenges in high school students. By statistically analyzing academic performance records, this work seeks to identify early indicators of behavioral health challenges, thereby empowering school teachers and administrators to promptly identify and support students requiring access to additional school resources. Our model uses declining and fluctuating academic performance data as a marker for the onset of such challenges as depression, anxiety, and bipolar disorders. At the heart of our approach lies the development of a machine learning model composed of two essential components: a deep encoder and a classifier. The deep encoder performs feature extraction from time series data consisting of grades received by students in various course subjects. The encoder identifies and highlights inherent patterns within these sequences that hold key information necessary for the early detection of behavioral health challenges. The output embeddings from the encoder are then fed into the classifier to label the students into one of three groups - stable, monitor, or critical. This approach has achieved a remarkable 93% weighted F1 score on our test set.