Defects are an important indicator of project quality; moreover, eliminating defects is a key objective of project management. Therefore, using the appropriate analytical tools and methods, training and testing the defect data, and selecting the best algorithm for the defect feature are important. These steps can directly reveal the decision rules for each defect, and they can assist in determining key approaches to construction site management for effective defect prevention. In this study, a model obtained by using the chi-squared automatic interaction detection (CHAID) algorithm was validated, and its prediction benefits were calculated. A total of 499 defect types were retrieved from the Public Construction Management Information System in Taiwan and used as the foundation of a statistical analysis of 990 construction projects with 17,648 construction defects. First, a cluster analysis of inspection scores and defect frequencies was performed to reclassify and establish a new grade. Next, five rules were established for using the decision tree to classify defects and inspection grades. Finally, results revealed that the prediction accuracy of the CHAID algorithm was 75.45%. The five rules can be used for defect management and prevention strategies.