This project investigates the comparative effectiveness of single-task learning (STL) and multi-task learning (MTL) approaches in predicting generalised symptom severity among individuals with schizophrenia using smartphone-based digital phenotyping data from the CrossCheck dataset.
While multi-task learning is commonly assumed to improve predictive performance by leveraging shared representations across tasks, this study critically evaluates whether that assumption holds in real-world mental health data, which is often sparse and highly personalised.
The objective of this research is to determine whether single-task models can outperform multi-task models under such conditions.
The study employs both single-task and multi-task variants of the Least Absolute Shrinkage and Selection Operator (LASSO):
- Single-task model: LassoCV
- Multi-task model: MultiTaskLassoCV
Model performance was evaluated using Root Mean Square Error (RMSE) to measure prediction accuracy. The Wilcoxon signed-rank test was applied to assess the statistical significance of performance differences between models.
Results indicate that:
- Single-task learning models outperform multi-task learning models when predicting generalised symptom severity in sparse datasets.
- Multi-task learning may introduce noise rather than shared benefit when applied to highly personalised mental health data.
These findings challenge prevailing assumptions in digital mental health modelling and highlight the importance of model selection based on data characteristics rather than convention.
This work contributes to:
- Improved understanding of AI model performance in digital mental health applications
- Development of more accurate personalised symptom prediction tools
- Evidence-based decision-making in healthcare-focused machine learning systems
Olajumoke Akinremi
Olajumoke Akinremi
Data Scientist | Digital Health Researcher |Founder & Director (Digital Programmes), Havilah Grace Foundation
Havilah Grace Foundation is a non-profit organisation delivering data analytics training and digital research initiatives for social impact, alongside medical relief programmes supporting widows.