Traffic Congestion Detection By Mining GPS Data Suhas Prakash Kaklij Department of Information Technology Siddhant College of Engineering, Sudumbare Pune, India Email: kaklijsuhas@gmail.com Prof. Sonali Rangdale Department of Information Technology Siddhant College of Engineering, Sudumbare Pune, India Email:sonali_rangdale@rediffmail.com Abstract- GPS data is available in large amount, also for devices having GPS a large amount data is being gathered over time. The mining of this huge data is to assist in discovery of the locations which face regular traffic congestion. User will have prior knowledge of such locations which helps in deciding whether or not to opt for that route. Avoidance of such path will also assist in reduction of congestion in such locations. Also observed that work done till now in this field does not give very precise results. The reason behind this is the no proper algorithm are selected and distinguished between on road and off road traffic. To consider all this we proposed this system. This system will be applied over GPS data i.e. data coming from verity of devices like mobile phones, tablets etc. In the technique with this system, these GPS data will be first cauterized using the K-means clustering algorithm. The clusters obtained are filtered out. On further processing these clusters a mining method of Naive bayes algorithm is used for mining for traffic Congestion detection and prediction Keywords-K-Means,Traffic Congestion,Traffic Jam Predicion, I. INTRODUCTION Road network is biggest network widely used for Transportation .Each city has its Road network. Road are used to travel from one point to another destination point. It is used for daily transport not only for the people but for goods and many more. The biggest problem now a days people facing is Traffic Congestion. People are not able to complete their work due this traffic problem. Also we observed that traffic congestion is of dynamic nature, it is not static. Means traffic congestion is variable as time passes. In current IT world we have lots of traffic information available with us in different format. By using this we can get the flow of traffic information with respect to location and time. This traffic information is important not only for current status of traffic but it can helps to analyze and predict upcoming traffic patterns. We can collect such information by processing GPS data. With availability of 2G and 3G enabled GPS devices, huge datasets are being collected with an average error of 2- 15m [2]. Using many of correction strategies such as map- based correction given in [2], this error can further be decrease. This is real time data which gives an opportunity to mine the traffic patterns of particular location .We can analyze such data to get the traffic congestion patterns which in turn helps to detect the location where traffic congestion is possible .Also we can predict possible traffic congestion. A. Related Work There is a massive amount of work undertaking in field of analyzing traffic patterns. H. Inose et al. in 1967, as given in [11], proposed how traffic signals are work systematically. Its work proposes for the minimization of delay time of vehicles and allocating preferential offsets to the optimum tree in a road network. In 2002, Ashbrook et al., as given in [10], projected user substantial locations and user activities using GPS data. Their work divided city in to various clusters using K-means clustering which further resulted into a Markov Model. Thus, their work focusses on analyzing user GPS data to mine user- significant locations. As per 2010, Lipan et al. in [5], mined traffic patterns from GPS data gathered from public transport. Their work focuses on monitoring bus schedules. Association rules are made on clusters in which each cluster has its own average speed. In 2011 [2] Mandal K and his team used probe vehicle technique for traffic congestion monitoring, system as whole tries to monitor the traffic flow pattern and then detect the congestion. As given in [3], Yao et al. proposed a speed pattern model which guesses traffic conditions and speed pattern using machine learning. In 2013[1] Anand Gupta and his team proposed a framework for traffic congestion detection focusing more on algorithm which reduces conflict of data for traffic Jam and Traffic signal. These works have given significant and helpful results. However, to the best of the authors’ findings, not much emphasis has been given to detection and prediction of traffic congestion with appropriately handling of on road & off road data as well as the Conflict between the Traffic signal and Traffic Jam .Also no proper selection of mining and clustering algorithm B. Motivation Detecting traffic jam based on simple rules, such as using a probe vehicle technique, velocity-based approach, and fuzzy logic might not handle the problem stated previously with great effect due to the following reasons 1. Proper Clustering algorithms not selected. 2. Suitable mining methods are not used 3. Not able to segregate on road and off road traffic AVCOE, Sangamner iPGCON-2015 SPPU, Pune 24th & 25th March 2015 Fourth Post Graduate Conference Page 1 of 5