Accurate construction cost forecasting is vital for delivering climate-resilient infrastructure, especially in hazard-prone regions like Australia. Inaccurate estimates can lead to project delays, budget overruns, or underbuilt structures that fail under climate stress. This paper presents a hybrid machine learning approach that integrates climate resilience parameters, such as Bushfire Attack Level (BAL) ratings, wind classifications and soil reactivity with architectural and locational attributes to forecast housing construction costs. A dataset comprising 500 housing project records (20 real, 480 synthetic) was compiled to represent diverse design and hazard conditions. We developed and evaluated multiple regression models, including XGBoost, Random Forest, and Linear Regression. XGBoost achieved the best performance with a mean absolute error (MAE) of $3,463 AUD and an R2 of 0.9936. SHapley Additive exPlanations analysis was applied to interpret the model’s predictions, revealing that total floor area, bushfire risk, and wind classification were the most influential cost drivers. Sensitivity analysis confirmed strong positive correlations between floor area and total cost, while also highlighting significant cost premiums for high BAL zones and severe wind classifications. The proposed framework demonstrates both high accuracy and transparency, making it suitable for early-stage planning, climate adaptation budgeting, and compliance with evolving building codes. By explicitly incorporating hazard-specific variables, the approach offers a scalable and adaptable solution for diverse regions and hazard types. This research supports informed decision-making, cost-effective resource allocation, and sustainable construction practices in the face of increasing climate risks.