Assessing and Improving Housing Flexibility in Duhok City Using Spatial Configuration, Space Syntax, and Machine Learning Surrogate Modeling: A Case Study

This study evaluates and enhances contemporary housing flexibility in Duhok City, Iraq, using an integrated architectural and space syntax workflow augmented with a lightweight machine learning surrogate. Ten residential units on plots of about 150 m² were assessed in original layouts Hi and redesigned layouts Hi′, yielding 20 plan instances. Twelve percentage-based architectural indicators capture openness, circulation efficiency, service clustering, polyvalent use, multi-access potential, and expansion capacity. Space syntax metrics were computed in DepthmapX, including connectivity, integration, and visual step depth, and aggregated into a composite Syntax Flexibility score. Redesign interventions produced consistent gains, increasing connectivity by up to 44.4%, integration by 38.7%, and visual step depth by 43.0%, with a mean increase in flexibility of 16.9%. For rapid screening of design alternatives, a regression surrogate was trained to predict Syntax Flexibility from architectural indicators and evaluated using leave-one-house-out group cross-validation to prevent leakage between paired plans. The best model, SVR with an RBF kernel, achieved an MAE of 12.8 and an RMSE of 15.7, outperforming a fold-wise mean baseline with an MAE of 15.3 and an RMSE of 20.5. Permutation importance highlighted circulation efficiency and open-plan ratio as the dominant predictors. Overall, compact circulation cores, increased openness, and polyvalent multi-access spaces improve housing flexibility, while the surrogate provides conservative decision support for early-stage evaluation.