Predicting Traffic Accidents Severity using Collaborative ML on Blockchain

With an increasing number of accidents and inefficient resources to inform authorities and hospitals quickly, there is a rise in the number of death cases because of traffic accidents. As per a report [9], in India, a total of 151,113 deaths have been reported in 480,652 traffic accidents in 2019, resulting in an average of 17 deaths per hour! As we continue to make progress towards building smart cities, the expectations to predict the anomalies such as traffic accidents also rise. In this paper, we have built a highly secure and increasingly accurate accident prediction system, which we believe brings us one step closer to Intelligent Transportation Systems. Our system runs a trained Machine Learning algorithm and can be updated by participants collaboratively, which makes it a system that can potentially achieve the highest accuracy following secure protocol. Besides, data breaches and changes in models' parameters are prevented with the use of blockchain. To avoid malicious participants, we have designed an astute incentive mechanism. To validate the claims and our system's performance, we have further used a dataset from the UK govt website with information about accidents, vehicles, and casualties. We have experimented with various machine learning models as participants of the system hosted on a blockchain.