Journal of Agricultural, Life and Environmental Sciences ISSN:2233-8322(Print) 2508-870X(Online) 농업생명환경연구

It is important to predict road cracks efficiently in advance for effective road maintenance. The aim of this study was to develop and evaluate a road crack prediction model for Gangwon-do using Google Teachable Machine Learning (TML). The training data consisted of asphalt (38 cracked, 31 normal) and concrete (38 cracked, 22 normal) road photos categorized into four classes. Twenty road photos (five from each class) were used for testing. Analysis of the average performance of the trained model showed that it predicted the cracked state of asphalt and concrete roads with accuracies of 94.2% and 95%, respectively, and the normal state with accuracies of 98.6% and 91.4%, respectively. The results indicated that the TML-based road crack prediction model is highly applicable for predicting the crack status of roads in Gangwon-do. However, additional data must be obtained to improve the predictive performance of this model. The findings of this study are expected to assist nonspecialists without machine learning expertise in developing artificial intelligence models for predicting road cracks in practical field applications.