<p>Line-level software defect prediction is a significant research direction in the field of software quality assurance, aiming to accurately identify potentially defective lines of source code in software systems. In recent years, the GLANCE approach, based on syntactic shallow features, and the DeepLineDP approach based on semantic features, have demonstrated outstanding performance. Meanwhile, the SPLICE approach, which integrates these two approaches, has significantly enhanced the ability to predict defective lines. However, existing research has not considered incorporating the syntactic deep structural features of code lines and has not fully utilized features with strong defect-indication capabilities in the integration process. To address the above two limitations, we propose a novel line-level defect prediction approach called PRaFFLineDP, which prioritizes the ranking of features with strong defect-indication capabilities and integrates syntactic shallow features with deep structural features on this basis. Notably, syntactic deep structural features offer a significant advantage over semantic features in terms of low-cost acquisition. Therefore, while maintaining high prediction accuracy, our approach achieves low-cost code line-level defect prediction (CLDP). We conduct a comprehensive comparison with existing state-of-the-art (SOTA) approaches using 6 performance metrics. Experimental results on 9 open-source projects demonstrate that PRaFFLineDP outperforms the current SOTA approaches CLDP and shows competitive performance. This finding underscores the critical importance of feature fusion in future CLDP research for constructing more effective defect prediction approaches.</p>