An approach to evaluate a classification model to predict a construction object’s state

A classification model which is designed to predict technical conditions of the construction object must not make the mistakes in categorizing “unfit” for normal operation objects as “fit” as such mistakes could lead to the accidents with a wide range of severities. So, to guarantee the mode’s results are correct, the scientists are doing model’s evaluation by the metrics, which high score is the indicator of model’s ability does not make the mistakes. However, on the question which metrics to be use to assess the model is not received a single answer as the researches’ conclusions often contradict each other while recommending the metric. The goal of this study is to propose the approach on how to select the metric to assess a binary classification model for predicting the technical conditions of the construction object. To meet the goal in the research Matthews Correlation Coefficient formula and F-measure were described using maximized Youden index, which value is possible to obtain when model doesn’t make the mistakes when predicting negative instance. The results of this work will provide the scientist the decision’s support method which recommendation depends on the optimal cut-off point on the ROC curve, so it improves the accuracy of the received evaluation score.