International Journal of Scientific Research and Engineering Development

The constant congestion of traffic is one of the major problems faced by big cities today. With the rapid urbanization, increased car usage and growing traffic congestion, chaos is caused on the road within 2 to 3 minutes. Fixed time traffic light systems operate on schedules and cannot react to real time traffic and environmental conditions or emergencies, leading to longer travel times, increased fuel usage, increased pollution and inefficient traffic flow. This paper introduces Traffic-Brain, an adaptive traffic control framework based on Large Language Model (LLM) and Artificial Neural Network (ANN), which aims to improve intelligent traffic prediction and decision-making. The framework feeds an ANN traffic-state classifier with environmental and traffic data, including air pollution index, humidity, temperature, cloud cover, visibility, and whether a vehicle is an emergency vehicle, etc., and classifies the traffic state, which is then sent to an LLM reasoning layer (GPT-4.1-mini) to generate an explanation-based, context-aware signal-control recommendation. The accuracy of the Hybrid ANN+LLM classifier was tested on a simulated data set of 14,454 records of intersection-intervals, with an accuracy of 99.55% (weighted F1-score 0.9955) and the end-to-end ANN+LLM decision pipeline reproduced the correct signal decision in 99.55% of the cases. Considered on a conservative, traffic-level-weighted perspective of the same data, Traffic-Brain achieves a 4.43% efficiency improvement across the network as a whole, and per-event efficiency improvements of up to 50% when subjected to HIGH traffic and up to 100% under emergency-vehicle traffic conditions. The results demonstrate that Traffic-Brain enriches the efficiency of the signals, decreases congestion, prioritises emergency vehicles correctly and is lightweight enough to be used in education and on a single-GPU system. The framework will be expanded to include realistically simulating a multiintersection network, integrating IoT sensors and coordinating between city scales in future work.

Gourav, Amandeep, "Design and Analysis of Traffic Brain: An LLM-Based Adaptive Traffic Control System" International Journal of Scientific Research and Engineering Development, V9(4): Page(644-649) May-June 2026. ISSN: 2581-7175. www.ijsred.com. Published by Scientific and Academic Research Publishing.