Time Series Forecasting with ML & DL

Research at DataBoss Inc. under Prof. S. Serdar Kozat on time-series analysis and forecasting, applied to the renewable-energy and finance sectors.

Contributions

• Developed a gradient-boosting variant that integrates error terms as features for forecasting renewable-energy generation, leading to publications in IEEE Signal Processing Letters and Digital Signal Processing.
• Built feature-selection pipelines (AFS-BM, Binary Feature Mask Optimization) that improved sequential-prediction performance through adaptive masking, with results published in Neural Computing and Applications and SIViP.
• Applied SCADA-driven forecasting and lag/rolling-window feature engineering for wind-energy generation.

This work also placed 4th in the Royal Dutch Shell Cashflow Forecasting Datathon (May 2023, ~1000 participants).