Item-level machine learning approach to identify influential predictors in self-report mental health scales

Hui, Le Yun (2025) Item-level machine learning approach to identify influential predictors in self-report mental health scales. Final Year Project, UTAR.

PDF Download (853Kb)

Abstract

This research introduces a innovative Long to Short approach on the DASS-42 mental health assessment tool for assessing stress levels among adults using machine learning. The data first retrieved for the mental heath assessment from Kaggle. The sum of the scores then obtained based on participants’ answers to every items in the complete questionairre. Next, feature selection techniques were applied to identify a selected items from the assessment based on participants’ responses, aiming to accurately predict outcome. Machine learning models were trained to get the smallest set of items required to reach a prediction accuracy of 95%. This study found that just three items are sufficient to predict stress status with at least 95 % accuracy compared to the full-scale assessment, using XGBoost and MLP model. However, demographic data such as age, gender, education level, and cultural background were not included in the analysis. The exclusion of these variables may limit the generalizability of the results, as demographic factors can influence how individuals respond to psychological assessments. Keywords: machine learning, DASS-42, stress assessment, feature selection, MLP, mental health screening Subject Area: QA76.75-76.765

Actions (login required)