HYBRID ANFIS-GA-RL TRAFFIC SIGNAL CONTROL: A SIMULATION STUDY OF NIGERIAN CITIES

Abstract

Rapid urbanisation in developing economies has rendered conventional and semi-adaptive traffic signal strategies increasingly ineffective against the non-linear, stochastic vehicular flows that characterise fast-growing megacities. This paper presents ITOS-v2, a novel Hybrid Intelligent Traffic Optimisation System that tightly integrates an Adaptive Neuro-Fuzzy Inference System (ANFIS) for interpretable fuzzy reasoning, a Genetic Algorithm (GA) for global parameter tuning, and a Reinforcement Learning (RL) actor-critic module for continuous online policy refinement. The tri-hybrid ANFIS–GA–RL architecture is evaluated through a calibrated MATLAB R2023b microscopic simulation of a single representative four-arm signalised intersection, using the U.S. Department of Transportation's Traffic Signal Performance Measures (TSPM) database (Data.gov, 2023) as the base signal-timing dataset, with input demand profiles and fleet-composition parameters rescaled to published contextual data for three Nigerian metropolitan clusters: Kano, Kaduna, and Lagos cities representing extreme to moderate congestion regimes with saturation degrees of 1.38–2.64. No field data were collected from Nigerian roads; the three cities are represented only by pairing each city's input demand volume with the results of a single base-case simulated intersection (Section 5.3), not by independently re-simulating that intersection under each city's full demand and saturation conditions or by simulating multiple real intersections. System performance is benchmarked against five increasingly complex baselines: Fuzzy Logic (FL), Artificial Neural Network (ANN), standalone ANFIS, standalone GA, and the ANFIS–GA hybrid. Evaluation across eight key performance indicators (KPIs) travel time (TT), average vehicle speed (AVS), throughput (TP), traffic density (TD), queue length (QL), delay time (DT), intersection delay (ID), and computational time (CT) reveals that ITOS-v2 achieves a 64.9% overall improvement over the fuzzy baseline, reducing TT to 0.20 h, ID to 0.01 h, DT to 0.50 h, and raising TP to 150 veh/h at minimum CT (0.0111 h). The base single-intersection model ranking is presented alongside each city's respective demand volume as contextual scaling information, rather than as evidence from independent re-simulations under each city's traffic conditions; city-specific re-simulation is identified as necessary future work. These findings position ITOS-v2 as a promising, computationally efficient foundation for next-generation smart-city traffic management, with field validation and network-scale extension identified as necessary next steps toward real-world deployment.