FeduLPM: Federated Unsupervised Learning-Based Predictive Model for Speed Control in Customizable Automotive Variants

Over speeding of vehicles is considered one of the most important factors for occurring road accidents. An advanced sensor technology motivates the researchers to move toward automatic vehicles, whereas speed decisions were taken based on the information collected from the sensor. Beyond its success, the expensiveness, and implementation failure in bad road conditions make it less practicable. In this work, a novel federated unsupervised learning-based predictive model (FeduLPM) technique has been proposed for controlling the speed in customizable automotive variants using data analysis from the location of the accident. Instead of sharing the whole accident data on a Heroku server to train the model, a pre-trained random forest algorithm is designed on each nearby location and the accident data is updated on the local model. Attributes such as weather, automotive variants, time, location, and zone geometry are taken for providing an exact speed limit. Finally, trained parameters of the local model are aggregated; then a global model is generated and distributed to local clients without sharing local data. The experimental setup is implemented using a Python-based machine learning (ML) algorithm that draws information from the vehicle based on the gathered attributes. The proposed FeduLPM achieved 96.7% of accuracy while processing with the different locations of Bengaluru city. From the experimental results, the proposed FeduLPM is a better assistant to bike and car drivers by providing an optimum solution to prevent accidents.