Traffic congestion is a persistent problem in many cities worldwide, creating a strong need for efficient models to predict traffic conditions across large transportation networks. Most previous studies have relied on single-task learning-based machine learning models, which often struggle with accuracy and involve high computational costs. In this paper, we propose a novel multi-task learning-based neural network framework to forecast traffic congestion large-scale transportation networks. The proposed model utilizes “hard parameter tuning with ReLU” to achieve improved performance across various road segments. Experimental results demonstrate that the proposed framework has outperformed state-of-the-art algorithms in terms of both accuracy and efficiency, making it a promising solution for traffic prediction in urban environments.