A Machine Learning and IoT-Enabled Framework for Shared Mobility in India: Reducing Traffic Congestion and Fuel Consumption through Data-Driven Ride Matching and Region-wise Impact Prediction

Abstract

India's road-vehicle fleet is growing at nearly 10% per year, far outpacing road-network expansion, while domestic crude-oil production has fallen and import dependency has climbed to almost 89%. In simple terms, more private cars are chasing the same limited road space and an increasingly imported fuel supply, and no existing tool tells a city, in advance, how much a shared-mobility policy would actually help. This paper addresses that problem by proposing a machine learning (ML) and Internet-of-Things (IoT) enabled framework with two cooperating modules: (i) an ML-based ride-matching engine that groups commuters with overlapping routes to minimise detour distance and split fares, and (ii) a region-wise traffic-and-fuel-savings prediction engine that combines the Greenshields speed-density model with an ML regression surrogate, calibrated on real-world natural experiments, to forecast congestion reduction and fuel savings for a given adoption rate. The framework is evaluated through an illustrative corridor-level simulation calibrated to Bengaluru's 2024–25 traffic conditions: at a 50% carpool-adoption rate with an average occupancy of three passengers per vehicle, the model predicts a 33.3% reduction in on-road vehicles, a 66.7% gain in average speed, and a 22.5% reduction in fuel consumption per kilometre. The paper also proposes a system architecture, implementation pipeline, and a set of publicly available Indian datasets (Vahan, PPAC, data.gov.in, Kaggle) on which the framework can be trained and empirically validated.