A supervised road traffic accidents detection with location extraction from microblogs

Abstract

Microblogging is a very popular way of sharing information about what is happening near realtime. Microblogs have been used to report about individuals daily life, advertisement or real world events as they are happening. This data is unstructured, voluminous with constantly changing topics that needs to be analyzed to make sense out of, in a most effective and efficient way which can be used during emergency cases. Large scale events like disasters or planned events like a head of state visiting another country, will normally generate a lot of tweets making it much simpler to detect. Small scale events like traffic updates, fires, road traffic accidents do not receive a lot of attention from microblog users hence more difficult to detect. Firstly reported on microblogs by road users, road traffic accidents are common in our roads which cause death and traffic congestion if not responded to in a timely manner. In this paper, text mining through machine learning and NLP techniques was proposed to detect road traffic accidents which includes, pedestrian, motorcycle, vehicles and other accident incidents on the roads or highways. In text classification for accident related tweets, SVM was adopted yielding a precision of 90.2% and a recall of 89.1 %. Location extraction from content proposed method, used syntactic analysis through Noun phrases and N-gram based matching with integration to local and external location databases giving a precision of 71.8% and a recall of 83.9%. An application called Traffic Accident Detection System (TADS) was developed based on the proposed methods where users can locate accidents in real-time by querying the system, ultimately visualizing the location of the event on the map for Nairobi, Kenya.