PERBANDINGAN KINERJA K-NEAREST NEIGHBORS DAN CONVOLUTIONAL NEURAL NETWORK UNTUK KLASIFIKASI CITRA KONDISI PERMUKAAN JALAN

Improving road infrastructure quality is an important aspect of transportation development and road user safety. Automatically assessing road surface conditions can accelerate maintenance and repair efforts. This study compares two classification methods, K-Nearest Neighbors (KNN) and Convolutional Neural Network (CNN), to evaluate road surface conditions based on digital images. Texture features are extracted using the Gray Level Co-occurrence Matrix (GLCM), including Contrast, Homogeneity, Energy, and others, to enhance the classification accuracy in KNN, while feature extraction and classification in CNN are performed automatically. The dataset used in this research consists of 1500 images of road surfaces with three different conditions: smooth, cracked, and potholes. Each condition contains 500 images with a resolution of 300x300 pixels. The results show that the KNN algorithm achieves an accuracy of 57.2%, while CNN demonstrates the best performance with an accuracy of 93.8%. for 80% training data and 20% testing data