Final year B.Tech project.
Team Members:
ENG20CT0002 - Abhijeet Sahoo
ENG20CT0018 - Prajjwal Mishra
ENG20CT0024 - Sathvik P
ENG20CT0027 - Shubham Raj Kashyap
According to the latest statistics disseminated by the World Health Organization (WHO), the global incidence of traffic accidents constitutes a profoundly disconcerting issue. These catastrophic events claim the lives of a staggering 1.2 million individuals annually, while an additional 50 million people suffer injuries. This grim toll translates to an approximate daily average of 3,300 fatalities and 137,000 injuries, posing a dire threat to both human lives and property safety. In tandem with these human costs, the direct economic repercussions of these accidents amount to a staggering $43 billion. Within the realm of traffic safety research, road accident prediction stands as a paramount focal point. The manifestation of road traffic accidents is intricately influenced by multifarious factors, encompassing the geometric attributes of road infrastructure, traffic patterns, driver characteristics, and the environmental conditions of the road. A multitude of studies have been conducted to prognosticate accident frequencies and scrutinize the underlying traits of traffic accidents. These investigations encompass endeavors such as the identification of hazardous locations or hotspots, analyses of accident injury severities, and the exploration of accident duration patterns. Some inquiries delve into the underlying mechanisms governing accidents, with consideration of additional factors like weather and road lighting conditions. Among the notable contributions to this domain, Lee et al. have pioneered the development of a probabilistic model that establishes a connection between significant crash precursors and alterations in crash likelihood. Furthermore, Abdel has constructed a previous crash prediction model through the adept utilization of the matched case-control logistic regression technique. Remarkably, a specific approach tailored to empower traffic authorities to forecast accident-prone areas at precise times remains conspicuously absent. The prediction of traffic accidents assumes a pivotal role in the comprehensive planning and administration of traffic systems, a task exacerbated by the inherent randomness and nonlinearity inherent in the elements governing traffic accidents. These encompass the intricate interplay of human behavior, vehicular dynamics, road infrastructure, climatic variables, and other intricate factors. The conventional linear analytical methods have proven inadequate due to limitations such as insufficient data and susceptibility to noise interference, ultimately yielding unsatisfactory predictive outcomes. Given these impediments, the conventional Backpropagation (BP) neural network emerges as an imperfect solution plagued by issues such as convergence to local minima, protracted training periods, and suboptimal accuracy. In contrast, the proposed model, whose details are beyond the scope of this abstract, demonstrably outperforms the BP network with an impressive 7.8% increase in predictive accuracy. This innovative approach presents a promising avenue for advancing our understanding of and response to the complex challenge of traffic accident prediction.
Keywords: Traffic accidents,Road accident prediction,Probabilistic model,Logistic regression,Neural network .