p05 / Conference
Federated Learning for Early Dropout Prediction on Healthy Ageing Applications
Fog and Mobile Edge Computing (FMEC 2023), 135-142 · 2023
In plain language
This work predicts early disengagement from healthy-ageing applications through collaborative training without transferring individual records.
Research context
Earlier dropout predictions could help application operators identify users who may need additional engagement support.
Approach
Cross-device and cross-silo federated scenarios use real data with unequal client distributions, class imbalance, and ambiguous labels.
Findings
Data selection and imbalance handling improve predictive accuracy, producing performance comparable to or better than conventional machine-learning models in the evaluated settings.
Summary based on abstract.