Abstract—Congestion Control (CC) is essential to ensure efficient network performance, especially for real-time applications where the stability of latency and throughput is critical. Traditional CC methods have difficulty adapting to the rapid fluctuations in the network, which can lead to performance degradation in a dynamic environment. Self-Clocked Rate Adaptation for Multimedia (SCReAM) congestion control algorithm has been proposed for efficient multimedia transmission, but it can still hardly adapt to challenging unpredictable congestion patterns. To address this issue, we propose a framework called LoSCM (Logistic Regression with SCReAM via Machine Learning). This machine learning-based congestion control mechanism integrates logistic regression with SCReAM to improve congestion prediction and response. LoSCM operates using a structured framework consisting of four major blocks. Extensive experimental evaluation demonstrates that LoSCM outperforms SCReAM in mitigating congestion and improving network adaptability. LoSCM can reduce congestion markers by 100%, Network Queue Delays (NQDs) by 17%, and Smoothed Round-Trip Times (sRTTs) by 2%. In addition, we also illustrate that the effectiveness of LoSCM in optimizing congestion control can reach a 44% reduction in NQD and an 11% improvement in sRTT. These results highlight that LoSCMs are efficient and robust mechanisms that can improve network performance in various real-time communication scenarios.