Software failure time series prediction with RBF, GRNN, and LSTM neural networks
Vitaliy Yakovyna, Nataliya Shakhovska
Abstract
The important task of software quality assurance is failure prediction. Time series forecasting methods can be successfully used for this purpose. This paper aims to study and compare the effectiveness of software failure prediction using different types of neural networks, in particular RBF neural network and recurrent neural networks (GRNN and LSTM). The Chromium browser failure dataset was used in this study. The dataset contains information about 11,001 bug reports for 759 consecutive days. Predictions were performed for 150-day, 500-day, and 759-day time series. Long time series of software failures cannot be approximated with satisfactory accuracy by an RBF neural network. Both GRNN and LSTM neural networks were able to predict the long software failures time series with RMSE values of 0.321 and 0.241 correspondingly.