<p>Recent advances in computer vision have led to significant progress in tracking and pose estimation of sports players. However, fewer studies have focused on behavior prediction using pose estimation in sports. In particular, predicting soccer fouls remains challenging due to the smaller image size of each player and the difficulty of incorporating information, such as the ball and player poses. In this research, we investigate a deep learning approach for predicting football fouls by integrating video data, bounding box locations, image details, and pose information to create a novel football foul dataset. Our model utilizes a combination of convolutional and recurrent neural networks (CNNs and RNNs) to effectively fuse these four modalities. Experimental results show that the full model outperforms the ablated versions, and that the RNN module, bounding box positions, image details, and pose information all contribute to improved foul prediction. Our findings offer an important reference for understanding football fouls.</p>