This thesis proposes a novel, integrated approach for enhancing energy efficiency in residential buildings by combining advanced machine learning techniques with visual network analytics. The research addresses limitations of current energy performance assessments, which often rely on static, expert-driven analysis and fail to capture complex, multi-domain feature interactions. Using a comprehensive UK EPC dataset (2008-2023), the study is structured into three core phases: (1) robust feature selection and ranking, (2) predictive modelling, and (3) network-based visual interpretability. In the first phase, multiple feature selection strategies including filter (mutual information, correlation), wrapper (recursive feature elimination), and embedded (LASSO, Random Forest) are systematically applied to identify the most impactful features influencing energy efficiency, such as CO2 emissions per floor area (CEPFA), total floor area (TFA), and heating cost (HCC). The second phase benchmarks various predictive models, demonstrating that ensemble methods (Random Forest, Gradient Boosting) outperform baseline approaches (Linear Regression, SVR, KNN) in both error reduction and predictive stability. The third phase introduces an innovative network visualisation framework that integrates feature importance and interaction values onto a graph structure, allowing clear identification of key feature “hubs” and synergies. This network visual approach enables a deeper understanding of how combinations of building characteristics, energy consumption, and operational factors jointly affect energy performance offering new, actionable insights for targeted retrofit interventions. This is the first work to seamlessly integrate consensus-driven feature selection, class-aware ensemble modelling, and multi-layer network explainability within a single, reproducible pipeline for building energy analytics. The framework not only achieves superior predictive accuracy, but also translates complex model outputs into intuitive, stakeholder friendly visual tools. The results provide actionable guidance for policymakers, energy consultants, and building managers aiming to optimise retrofit strategies and achieve sustainability goals. Future work is outlined to address scalability, dynamic data integration, and adaptation across building types and regions.