Performance Prediction Methodology Based On Machine Learning

This paper intends to apply the theory of machine learning to the performance prediction of complex application system, and proposes a method for constructing a performance prediction module based on Machine learning. This method can be used in many types of applications as a common, loose coupled and embedded one, and the cost and risk will be reduced by using it. In this method, a training set is collected during the process of the performance test.Then a classifier is trained using machine learning method. Finally, the trained classifier is packaged into a prediction module embedded into the system, which could be used to predict various performance factors, such as the responsive time of users' operations. At the same time, this module can also update itself by using the newest data in order to adapt to the changing environment.This paper takes Naive Bayes Classifier as an illustration to describe how to construct a prediction module in detail. Compared to traditional methods, Naive Bayes method shows a variety of superiorities, such as high accuracy, high efficiency, strong robustness and loose couple. A comparative experiment based on a financial report system shows that, our method's accuracy rate reaches 65% and the time spent during the training process is less than that of traditional methods.Furthermore, this paper extends the prediction method on the basis of Bayes module. On depth, Attribute Weighed method and Simulated Annealing Genetic algorithm are proposed to enhance the accuracy and environment adapting ability. On extent, many other algorithms in machine learning are selected to construct prediction module instead of Naive Bayes to enhance the expandability and particularity.