Optimizing Video Streaming Quality with DBN-Based ABR Prediction Models
Bello Usman Sani
Abstract
The rising demand for seamless and high-quality video streaming has intensified the need for intelligent Adaptive Bitrate (ABR) algorithms that can dynamically respond to fluctuating network conditions and user preferences. Traditional ABR techniques often rely on rule-based logic, which lacks the flexibility to handle complex and real-time network scenarios, leading to frequent buffering and degraded Quality of Experience (QoE). This research work aims to compare the model’s performance to existing ABR methods (CNN) by evaluating metrics such as buffering time, video quality, accuracy, precision, recall, and F1-score to measure improvements in streaming quality and user satisfaction. The scope of the study will involve the design of the DBN architecture, optimization of its hyperparameters for improved prediction accuracy, and the integration of the model into a streaming framework. A rich dataset encompassing network latency, bandwidth, packet loss, user behavior, and video playback characteristics was preprocessed, normalized, and balanced using the Synthetic Minority Oversampling Technique (SMOTE). The DBN and CNN models were trained to predict optimal bitrate transitions, aiming to reduce rebuffering and enhance video quality. Evaluation was conducted using key performance metrics, including accuracy, precision, recall, F1-score, buffering time, and video quality. Results show that the DBN model outperformed CNN, achieving 93% accuracy, 94% precision, 92% recall, and 93% F1-score, alongside reduced buffering time and improved video quality consistency. These findings demonstrate the effectiveness of DBNs in delivering robust and adaptive streaming performance and highlight their potential as a scalable solution for future intelligent ABR systems.
References