At present, the evaluation of the comprehensive performance of urban office buildings remains an area of significant discussion. This research aims to optimize the building performance of office buildings in the hot summer and warm winter (HSWW) region, focusing on three key aspects: energy use intensity (EUI), useful daylight illuminance (UDI), and percentage of thermal comfort (PTC). The study employs the Hyperparameter Optimization (Hyperopt)-Categorical Boosting (CatBoost)-Strength Pareto Evolutionary Algorithm 2 (SPEA2) multi-objective optimization method, generating 3,000 datasets via Latin Hypercube Sampling (LHS). Building performance parameters are simulated using the Ladybug and Honeybee models, and energy consumption and comfort levels are predicted using the CatBoost model. Subsequently, Hyperopt is used to optimize hyperparameters, and the SPEA2 algorithm is applied to identify Pareto optimal solutions. The results indicate that Hyperopt-CatBoost demonstrates excellent predictive performance, with R² values of 0.996, 0.954, and 0.985 for energy consumption, lighting, and thermal comfort, respectively. By using the SPEA2 multi-objective optimization (MOO) algorithm to optimize design parameters, energy consumption is reduced by 29.61%, lighting efficiency improves by 59.61%, and comfort increases by 37.69% compared to the original design. This study provides a systematic optimization plan and data support for energy-saving design, improving comfort, and enhancing lighting efficiency for office building renovation in urban villages.