Developing a machine learning-based building repair time estimation model considering weight assigning methods

Abstract Recently, the maintenance of aged buildings has gained significant attention, particularly with the increase in deteriorating buildings worldwide. The degradation of buildings causes several problems in terms of safety, structural, functional, and economic aspects. Thus, predicting the building repair time is an essential first step to cope with maintenance-related problems. In particular, globally, residential buildings in highly populated areas have accounted for a large portion of building maintenance or repair. Thus, this research developed a model for predicting the repair time for the building type by applying the genetic algorithm (GA), multiple linear regression analysis (MLR), feature counting method, and fuzzy-analytical hierarchy process to case-based reasoning. An experiment was conducted to validate the feasibility of the developed model using 13 randomly selected test cases. The results obtained from this experiment validated the estimation performance of the four weighting methods. The case similarity of the retrieved cases was approximately 90%, implying that cases similar to the test cases were extracted from the database. The mean absolute error ratios of the repair time determined by the 1-, 5-, 7-, and 10-nearest neighbors were typically less than approximately 10%, thereby proving the applicability of the developed model. This research also demonstrated that the GA and MLR approaches outperformed the other methods. This study contributes to an understanding of building management by not only suggesting a systematic approach for estimating the repair time of residential buildings, but also by demonstrating the effect that different weighting methods have on the estimation performance using case-based reasoning.