Simplified Deep Forest Model Based Just-in-Time Defect Prediction for Android Mobile Apps
Abstract : The popularity of mobile devices has led to an explosive growth in the number of mobile apps in which Android mobile apps are the mainstream. Android mobile apps usually undergo frequent update due to new requirements proposed by users. Just-In-Time (JIT) defect prediction is appropriate for this scenario for quality assurance because it can provide timely feedback by determining whether a new code commit will introduce defects into the apps. As defect prediction performance usually relies on the quality of the data representation and the used classification model, in this work, we modify a state-of-the-art model, called Simplified Deep Forest (SDF) to conduct JIT defect prediction for Android mobile apps. This method uses a cascade structure with ensemble forests for representation learning and classification. We conduct experiments on 10 Android mobile apps and experimental results show that SDF performs significantly better than comparative methods in terms of three performance indicators.
? The tree-based models are trained to predict the code by generating new nodes taking into account the existing tree structure.
? To prepare a labeled dataset, the authors use the existing static bug detection software to identify the specific kind of bugs.
? The existing comparative studies such as show that while the state-of-the-art deep learning techniques usually perform better than standard deep learning and traditional metrics-based ones.
? We hope that this survey can be useful for researchers and practitioners in the software defect prediction, code understanding and other related fields.
? The main idea is that an artificial neural network with multiple layers is capable of progressively extracting the higher-level features from the original data to solve complex problems.
? To deal with the problem that of the shallow machine learning based software defect prediction model can not deeply mine the software tool data, we propose software defect prediction model based on improved deep forest and autoencoder by forest.
? Then, the improved deep forest algorithm and eForest are applied to software defect prediction problem.
? Software defect prediction is a binary classification problem, which is to analyze the quality of software modules.
? To solve the problem that multi-grained scanning in deep forests may lose important information, data augmentation method is used to transform the original input features.
• Developing a fault prediction model is a challenging problem, and many approaches have been proposed throughout history.
• The proposed approach represents the code as a sequence of code tokens, which is fed into a LSTM system to transform code into a feature vector and a token state representing the semantic information of the token.
• In, the features learning technique based on CNN is proposed. This model extract features from token vectors in the AST of the code and learns the transferable joint features.
• In, a new deep forest model is proposed for the software defect prediction. To detect the essential defect features, it uses the cascade learning strategy, which consists in reforming a set of the random forest classifiers into a layered network.
? Software defect prediction is an important way to make full use of software test resources and improve software performance.
? Moreover, contrast to DNNs, the gcForest has much fewer hyper-parameters and its performance is robust in different hyper-parameter settings.
? We use data augmentation algorithms to increase the amount of data, challenging the time and efficiency of our algorithm.
? It shows that our algorithm has obvious advantages in efficiency compared to other deep learning algorithms.
? Moreover, there are mainly three techniques are used for implementing the software detect prediction models, as classification, regression and clustering.
|