Owing to the stigmatization of mental illnesses such as depression in India [1], there is a need for indirect unsuspecting ways to identify risk for depression and provide timely intervention. Healthy-Unhealthy Music scale (HUMS) [2] is one such assessment tool developed on Australian population that uses music engagement as an indicator of anxiety levels and potential high-risk for depression as assessed by Kessler’s Psychological Distress Scale (K10). The current study aims to ascertain its validity in an Indian setting followed by applying machine learning approaches to predict mental well-being from music associations. A diverse group comprising Indian adult population was assessed using HUMS and mental well-being and proneness to depression measures. HUMS structure investigated via Exploratory factor analyses, and concurrent validity tested with correlations to depression risk and wellbeing revealed high external validity and applicability of HUMS in Indian adult population. Furthermore, very low in-sample error for models like Support Vector Machines (SVM) with nonlinear kernels suggests an underlying pattern between HUMS responses and K10 score. Finally, a two-class model resulted in out of sample accuracy of 81%. To conclude, HUMS demonstrates high generalizability and hence applicability in Indian adult population and potential for employing ML models to capture the underlying pattern.
Cite as: Agarwal, R., Singh, R., Saarikallio, S., McFerran, K., Alluri, V. (2019) Mining Mental States using Music Associations. Proc. Workshop on Speech, Music and Mind (SMM 2019), 56-59, doi: 10.21437/SMM.2019-12
@inproceedings{agarwal19_smm,
author={Rajat Agarwal and Ravinder Singh and Suvi Saarikallio and Katrina McFerran and Vinoo Alluri},
title={{Mining Mental States using Music Associations}},
year=2019,
booktitle={Proc. Workshop on Speech, Music and Mind (SMM 2019)},
pages={56--59},
doi={10.21437/SMM.2019-12}
}