PURPOSES : The purpose of this study is to enhance the reliability of artificial intelligence for a noise-based pavement condition rating system (to a target performance of 95 %). METHODS : By comparing four types of pattern recognition artificial intelligence, this work acquires high-quality learning data and optimizes data learning through analysis of error characteristics. RESULTS : The system reliability improved up to 97 % (82 % in a prior study). In addition, 100 % was achieved for the E(F) condition grade, which has a direct impact on maintenance decision making. CONCLUSIONS : KNN-DTW (K-nearest neighbor dynamic time warping) is judged to be the most suitable type of artificial intelligence for a noise-based pavement condition rating system; a 4-grade system is the most suitable for classifying pavement condition.