Risk Analysis: Road Fatalities in India​

As part of my Risk Management class (IE590) at Purdue University, I collaborated with Harsh Agarwal to investigate the alarming rate of road traffic accidents, which claim 1.3 million lives annually, with India accounting for 11% of these fatalities. The 18-45 age group, which represents 67% of total fatalities, is particularly affected, leading to significant socio-economic losses as these individuals are often primary earners. Despite the implementation of various road safety measures, issues such as inconsistent enforcement of traffic laws, resistance to behavioral change, and unreliable accident data have hindered effective intervention.

To address these challenges, I applied Bayesian theory and utilized Netica to develop a structured risk analysis framework. The goal was to identify key risk factors for accidents, predict probabilities, and generate actionable insights. This data-driven approach aimed to provide a foundation for improving enforcement strategies, informing policy decisions, and ultimately reducing road traffic fatalities.

To model the relationships between various factors contributing to road accidents, I chose a Bayesian network as my approach.  To construct this network, I gathered data from the Indian government, particularly from the Ministry of Road Transport and Highways (MoRTH).  I focused on traffic accident statistics, road safety measures, and the reported causes of accidents.  This data included details on accidents across different regions, vehicle types involved, pedestrian fatalities, and information about road infrastructure and traffic enforcement.  I then used this data to build the probability model.  First, I identified the key risk factors and how they might be related.  For example, I considered how road type might influence accident severity.  Next, I applied Bayesian theory to calculate the conditional probabilities of different events, like the chance of an accident being fatal given specific road conditions or vehicle types.  I used Netica software to build, test, and refine the network, making sure it realistically represented accident scenarios in India and could offer useful insights for improving road safety. For building the Conditional Probability Tables (CPTs), I primarily used Maximum Likelihood Estimation from the MoRTH data.  Where data was limited, I incorporated expert knowledge and used smoothing techniques to handle cases with few observations.  I also performed sensitivity analysis and validation to ensure the network's accuracy and reliability.

The Bayesian network revealed key relationships within the analyzed road accident data, helping quantify the risk associated with each factor. The most striking observation was the strong association of National Highways with fatal accidents, suggesting that the higher speeds and traffic volumes on these roads contributed significantly to severe outcomes. While driver-related factors like overspeeding and wrong side driving were also prominent, their impact was likely amplified on uncontrolled strech of roads. The high proportion of two-wheeler involvement in accidents underscored their vulnerability, highlighting the need for targeted safety measures. The network helped quantify the elevated risk faced by two-wheeler riders. Overall, the network emphasized the complex interplay of road infrastructure, driver actions, age and vehicle type in determining accident outcomes, pointing towards the need for multifaceted interventions that addressed these interconnected elements.