METCN: A Multi-Task Enhanced TCN Model for Software Fault Detection and Correction Prediction

Although there are many software reliability growth models, they still face some challenges. Firstly, these models can be divided into two categories: parametric models and non-parametric models. The parametric models often make assumptions that differ from the actual situation, which affects their generalizability and performance. Secondly, although the non-parametric models based on deep learning have shown impressive performance, these models mainly focus on recurrent neural networks, which are prone to problems such as gradient explosion, gradient disappearance and long-term dependence. Thirdly, most deep learning models primarily focus on the software fault detection process (FDP) and overlook the software fault correction process (FCP), but these two processes are interrelated and crucial for improving software reliability. In this paper, we propose a Multi-task Enhanced TCN (METCN) model, which leverages multi-task learning and Temporal Convolutional Network (TCN) to capture temporal features of software failure data in FDP and FCP, thereby establishing a dual-process software reliability growth model. Moreover, the model integrates a Squeeze-and-Excitation (SE) module to increase attention to important feature channels. We present the model's fundamental principle and modeling framework, as well as performance comparison criteria considering dual processes. Finally, we validate the capability of the proposed model using two real failure datasets. The experimental results show that the proposed METCN model has better performance than other models based on Recurrent Neural Network(RNN), Long Short-Term Memory(LSTM), Gated Recurrent Unit (GRU) and TCN.