Presented during CBT4 at Aalto University, Azin Farahani’s July 2026 presentation for the HumanIC Project at Granlund Oy offers a concise guide to extracting actionable insights from large building datasets. The session emphasizes that high-quality, preprocessed data is the foundation of any analysis, highlighting the necessity of cleaning messy data and visualizing it to uncover hidden patterns before applying complex algorithms. Farahani explains how foundational statistics should precede predictive modeling, moving from analyzing historical behaviors—like overheating in apartments—to forecasting future system needs using supervised, unsupervised, and reinforcement learning techniques. Crucially, the presentation stresses that domain knowledge in building physics provides a significant advantage in feature engineering and that all predictive models must remain physically plausible and explainable. It concludes with strategies for managing massive datasets using high-performance computing when standard hardware falls short.
