Hybrid AI model for social network-based flow prediction

Abstract Accurate traffic flow prediction is a crucial aspect of intelligent transportation systems, yet it remains challenging due to disruptions caused by non-routine events such as accidents, road closures, and severe weather. Traditional statistical and machine learning models often fail to fully capture these anomalies, resulting in reduced prediction accuracy. This research proposes a hybrid approach that enhances traffic forecasting by integrating social media data, specifically tweets, to detect real-time events that may impact traffic conditions. The methodology consists of three key phases. First, social media data is collected and preprocessed to remove noise, spam, and irrelevant content, thereby reducing data complexity without sacrificing information quality. Second, a Hidden Markov model is applied to analyze time-series patterns involving tweet frequency, temporal factors, and weather conditions. Finally, a SincNet-based convolutional neural network is utilized for high-accuracy traffic prediction. SincNet’s ability to effectively process time-series data makes it particularly well-suited for modeling traffic behavior. The model demonstrates strong performance in forecasting morning traffic based on tweet data available up to midnight of the previous day. Experimental results show that the proposed approach achieves improved prediction accuracy of up to 90%, with 89% precision and a reduced error rate of 0.2, outperforming existing state-of-the-art models.