Making an informed decision requires an accurate prediction of building energy efficiency. However, it can be challenging due to the evolving and complex structure of buildings. Physical features, climate variations and system performance could be considered as factors that cause uncertainty, demanding accurate data collecting and advanced methods. Building performance simulation methods, are commonly classified into two main methods: physics-based and data-driven, which rely heavily on the quality and quantity of the input data. However, simulated data frequently are not accurate and don’t represent real-world conditions, providing an energy performance gap, emphasising the significance of accurate datasets. This study aims to improve the accuracy and reliability of datasets used in building energy performance analysis process by integrating physics- based simulations and data-driven methods. This method improves prediction models by analysing large datasets that cover various building characteristics and operational data. The study collects data from multiple databases and sets missing parameters using various building regulations to provide a valid foundation for machine learning (ML) models. Initially, a parametric archetype simulation model is created to simulate reliable data with advanced tools such as Rhinoceros (Rhino) for 3D modelling, Grasshopper for parametric algorithm generation, and Ladybug, Honeybee, and Energy Plus for energy and environmental analyses. Subsequently, a statistical analysis is used to identify and prioritise Key Performance Indicators (KPIs) that contribute to energy losses in buildings and assist us to make an efficient decision which leads to offering optimal retrofitting solutions. Finally, the proposed method applied to a semi-detached house in in London, UK, located in ASHRAE climate zone 4A (Mixed- Humid), as a case study and then data collection and analysis process expanded to encompass 500 scenarios across three additional UK climate zones (5A, 5C and 6A). This approach generated a comprehensive and reliable data set with simulated data assessing energy consumption and thermal comfort across various Window-to-Wall Ratios (WWR).