Traffic congestion and its associated pollution pose significant challenges for modern smart cities, primarily influenced by vehicle density, travel time, and variable speed. This paper introduces an improved multi-objective optimization framework designed to minimize air pollution, reduce travel time, and enhance overall traffic flow while ensuring low computational load and high prediction accuracy. To this end, an improved multi-task learning-based optimization algorithm (MTLBO) is proposed and validated using real-world traffic data from the Stadshouderskade street in Amsterdam, Netherlands. Comparative tests with benchmark algorithms, including the Best Random Average (BMR) and JAYA algorithms, demonstrate the superior performance of the proposed approach. The improved MTLBO algorithm achieves an average accuracy improvement of 93.77% over the BMR algorithm and 91.53% over the JAYA algorithm, with enhanced convergence stability and scalability. These results highlight the potential of the proposed model as a robust and efficient solution for sustainable traffic management in smart cities, which could be effective given its wide applicability in our country.