Assessing cloud QoS predictions using OWA in neural network methods

Area of Science:

• Computer Science
• Artificial Intelligence
• Cloud Computing

Background:

• Quality of Service (QoS) is crucial for evaluating service-oriented applications.
• Large, sparse QoS datasets pose challenges for accurate, efficient prediction.
• Dimensionality reduction is needed to manage complexity and improve predictive performance.

Purpose of the Study:

• To introduce a novel approach using an Induced Ordered Weighted Average (IOWA) layer for cloud QoS data prediction.
• To reduce the dimensionality of extensive QoS datasets without significant information loss.
• To enhance the accuracy and efficiency of predicting future QoS intervals.

Main Methods:

• Implementation of an IOWA layer within the prediction model.
• Evaluation of cloud QoS prediction using the IOWA operator combined with nine neural network architectures (e.g., LSTM, GRU).
• Benchmarking prediction accuracy using RMSE, MAE, and MAPE metrics on Amazon EC2 US-West instance data.

Main Results:

• The IOWA approach reduced the dataset size by 66% (from 2016 to 672 records).
• Prediction accuracy was improved or maintained compared to traditional methods.
• The method effectively handled complex, nonlinear predictions and reduced data dimensions.

Conclusions:

• The IOWA layer is an effective technique for dimensionality reduction in cloud QoS data.
• This approach offers a practical solution for managing large datasets and improving prediction accuracy.
• Stakeholders can better manage extensive QoS data and complex nonlinear predictions for service-oriented applications.