Three different pattern recognition methods were applied in this researchin order to classify freeway traffic flow conditions based on the characteristics of the flow. The applied methods are K-Means, Fuzzy C-Means, and CLARA, which fall into the category of unsupervised learning and require the least amount of knowledge about dataset. The classification results from the three clustering methods were compared with the Highway Capacity Manual (HCM) level-of-service criteria. Through this process, the best clustering method consistent with the HCM classification was identified. Clustering methods were then used to further categorize oversaturated flow condition to supplement the HCM classification. The clustering results supportedthe HCM's density-based level-of-service criteria for uncongested flow. Additionally, the methods provide a means of reasonably categorizing oversaturated flow conditions that the HCM is currently unable to do.
Abstract