Dynamic QoS Prediction Algorithm Based on Kalman Filter Modification.

Area of Science:

• Computer Science
• Software Engineering
• Artificial Intelligence

Background:

• Service-oriented architectures (SOA) present challenges due to numerous services with varying Quality of Service (QoS).
• Evaluating all available services for optimal QoS is often resource-intensive for users.
• Accurate QoS prediction is crucial for users to select appropriate services.

Purpose of the Study:

• To propose and validate the MFDK model, a novel QoS prediction algorithm for service-oriented architectures.
• To address the challenge of sparse historical QoS data and improve prediction accuracy.
• To enhance the selection of high-quality services by users.

Main Methods:

• The MFDK model utilizes non-negative matrix decomposition to handle sparse historical QoS data.
• A deep neural network is employed for predicting future QoS values.
• A Kalman filter algorithm refines predictions using real-time QoS observations.

Main Results:

• The MFDK model demonstrates superior prediction accuracy compared to baseline models on the WS-DREAM dataset.
• The model maintains robust prediction performance across varying tensor and observation densities.
• Ablation and parameter tuning experiments confirm the model's effectiveness and rationality.

Conclusions:

• The MFDK model offers a significant advancement in QoS prediction for service-oriented architectures.
• The hybrid approach effectively addresses data sparsity and improves prediction accuracy.
• This research provides a valuable tool for users seeking to optimize service selection based on QoS.