| Summary: | 碩士 === 國立臺灣科技大學 === 工業管理系 === 107 === Weather condition such as raining or snowing indeed influences the traffic condition. However, not every road has the similar impact from the same weather condition. Moreover, the varied road condition of developing countries urban road increases the diversity of road characteristic. Therefore, understanding the road characteristics that influence its weather sensitivity is important for traffic management or logistic industry. The Weather Sensitive Road (WSR) is identified as road that sensitive to weather. In this research, the one-year traffic congestion data generated from citizens of Jakarta’s smartphone was collected. The spatiotemporal data was captured in the form of traffic congestion speed at a particular time and on a certain location. There were two consecutive models used: K-means clustering model and the Random Forests prediction model. K-means clustering model was used to acknowledge the existing levels of weather sensitivity among Jakarta’s urban road. The Random Forests prediction model was chosen to identify the importance of road characteristics given its weather sensitivity level. The experiment was performed by identifying 4 clusters to 50 different clusters of weather sensitivity levels. The best number of clusters was chosen by the Pareto Front method. K-means performance measurements and the Random Forests performance measurement were used to consider the Pareto Front. The best set of Pareto Front is 4, 6, 11, 14, 16, 19, 21, 39, and 49 of weather sensitivity levels. The 11 clusters set was chosen to minimize the complexity while to maintain the fair number of clusters. It shows that the top 4 of important features are longitude, latitude, distance to the nearest mall, and elevation.
|
|---|
