Leveraging spatial data infrastructure for machine learning based building energy performance prediction.

Area of Science:

• Building energy performance (EPB) assessment and management.
• Spatial data infrastructure (SDI) and data modeling for buildings.
• Machine learning (ML) applications in energy efficiency.

Background:

• Buildings account for a third of energy consumption, with significant inefficiency in existing stock.
• Lack of EPB calculation and integration into national spatial data infrastructure hinders policy development.
• Türkiye's building sector faces challenges in energy efficiency and greenhouse gas reduction.

Purpose of the Study:

• To design and implement a building energy data model extending Türkiye's NSDI.
• To develop and validate ML models for predicting building energy performance scores.
• To support data-driven policy-making for energy efficiency and sustainability.

Main Methods:

• Designed a building energy data model as an extension of the national spatial data infrastructure (NSDI).
• Implemented the model using real data from energy performance certificates in Istanbul's Tuzla District.
• Developed and applied ML algorithms (RF, GBM, LightGBM, XGBoost) for energy performance prediction.

Main Results:

• The XGBoost model achieved high predictive accuracy with R² = 0.818.
• Performance metrics included RMSE = 5.153, MAE = 2.886, and MAPE = 3.369.
• The model demonstrated reliability in estimating building energy performance scores.

Conclusions:

• The developed model and ML predictions offer a reliable tool for assessing district-level EPB.
• Findings can inform energy efficiency roadmaps, legislation, and incentive programs.
• The study highlights the need for broader validation of the model across different regions.