Predict Traffic State Based on PCA-KMeans Clustering of neighbouring roads

During the past few years, time series models and neural network models have been widely used to predict traffic conditions based on historical data, speeds, weather, accidents, and special holidays. However, in previous studies, these models were commonly used for predicting traffic flow, rather than predicting traffic flow propagation. Research in traffic flow propagation is relevant because it may guide people in avoiding neighbouring roads which are affected by congestion. We proposed the similarity of Principal Component Analysis (PCA) to investigate the relationship between roads by clustering similarity values between target roads and neighbouring roads. The results were then visualized on a map for further observation. Furthermore, the high relationship roads obtained from the cluster were then used for predicting traffic state using a naïve Bayes method. Based on the visualization of results on maps, and by observing the prediction results using naïve Bayes, obtained that utilizing PCA with K-Means improves the outcomes in obtaining high relationship roads compared with k-means only.