Sea Hero Quest: Clinical Data Analysis
Large-scale data analysis of 4+ million players' spatial navigation patterns to identify cognitive biomarkers for dementia research, processing 78,000+ complete game sessions.
Overview
Analysis of raw gameplay data from Sea Hero Quest, a citizen science mobile game designed to collect spatial navigation data for dementia research. This project processes data from over 4 million players to identify patterns in navigation performance that could serve as early biomarkers for Alzheimer's disease. The analysis includes cross-level correlations, age-based performance analysis, and predictive modeling.
Technologies
Key Features
- ✓Processing 78,000+ complete game sessions across 74 levels
- ✓74×74 cross-correlation matrix for level performance analysis
- ✓Z-score categorization and percentile ranking systems
- ✓Quintile segmentation for player performance grouping
- ✓Predictive modeling achieving ~60% accuracy on later level performance
- ✓Age-based cognitive performance correlation analysis
Challenges
Processing millions of gameplay records required efficient big data techniques using Dask for parallel processing. Identifying meaningful patterns in noisy gameplay data while controlling for confounding variables like game difficulty and player demographics was a significant analytical challenge.
Results
Discovered that levels 43, 52, 53, 58 & 71 show the strongest correlation with overall cognitive performance, all sharing 'hard' difficulty and 'checkpoint' level types. Interestingly, age was NOT a significant predictor of performance, challenging common assumptions about cognitive decline.