← Back to Projects
Evaluating Federated Learning for Mood Inference from Smartphone Sensing Data
📅 Aug 2025 – Nov 2025
📚 Course Project
🏛️ IISER Bhopal
Course Project
Federated Learning
Affective Computing
Overview
Evaluated federated learning on the DiversityOne dataset (250 participants, 6 countries). Preprocessed
heterogeneous data using windowing, KNN-imputation, and feature extraction (50–63 features). Designed a
hybrid FL model with a shared global encoder and local country-specific encoders for personalized
cross-cultural mood inference. Achieved near-baseline AUROC and matched centralized performance across
countries.
Key Highlights
- Evaluated federated learning on the DiversityOne dataset (250 participants, 6 countries)
- Preprocessed heterogeneous data using windowing, KNN-imputation, and feature extraction (50–63 features)
- Designed a hybrid FL model with a shared global encoder and local country-specific encoders for personalized cross-cultural mood inference
- Achieved near-baseline AUROC and matched centralized performance across countries