Neetu Sardana

Improving Logging Prediction on Imbalanced Datasets: A Case Study on Open Source Java Projects

Logging is an important yet tough decision for OSS developers. Machine-learning models are useful in improving several steps of OSS development, including logging. Several recent studies propose machine-learning models to predict logged code construct. The prediction performances of these models are limited due to the class-imbalance problem since the number of logged code constructs is small as compared to non-logged code constructs. No previous study analyzes the class-imbalance problem for logged code construct prediction. The authors first analyze the performances of J48, RF, and SVM classifiers for catch-blocks and if-blocks logged code constructs prediction on imbalanced datasets. Second, the authors propose LogIm, an ensemble and threshold-based machine-learning model. Third, the authors evaluate the performance of LogIm on three open-source projects. On average, LogIm model improves the performance of baseline classifiers, J48, RF, and SVM, by 7.38%, 9.24%, and 4.6% for catch-blocks, and 12.11%, 14.95%, and 19.13% for if-blocks logging prediction.

Visheshagya: Time based expertise model for bug report assignment

The brisk escalation in scale of software systems has made bug triaging an imperative step in bug fixing process. A huge amount of bug reports is submitted daily to bug tracking repositories. Although this practice assists in building a reliable and error-free software product but handling a large amount of work becomes challenging. Bug assignment, an essential step in bug triaging, is the process of designating a suitable developer for the bug report who could make code changes in order to fix the bug. Various approaches ranging from semi to fully automatic bug assignment are proposed in literature. These approaches are mostly based on machine learning and information retrieval techniques. Since the information retrieval based activity profiling approach achieves higher accuracy, they are more often used in recent studies. Time factor based normalization in activity profiling could play a vital role in analyzing the level of expertise (or knowledge) of developers as the knowledge decays with time. This paper proposes a time oriented expertise model, Visheshagya, which utilizes the meta-fields of bug reports for developer selection. The proposed technique is used to prioritize the developers actively participating in software bug repository on the basis of their current knowledge. The proposed approach has been validated on two popular projects of Bugzilla repository, Mozilla and Eclipse. The result shows that time based activity profiling of developers outperforms existing information retrieval based bug report assignment and achieves an improvement of 14.3% and 9.95% in the accuracy of top-10 list size in Mozilla and Eclipse projects respectively.

Significance of Clustering Coefficient over Jaccard Index

Link prediction is a key task to identify the future links among existing non-connected members of a network, by measuring the proximity between nodes in a network. Node neighbourhood based link prediction techniques are immensely used for prediction of future links. These techniques can be applied on various applications like biological protein- protein interaction network, social network, information network and citation network to predict the future links. Every network has got certain structural properties. For predicting the accuracy of future links, structural properties might play an important role. Current work is an effort to find out the correlation between structural properties of network and prediction accuracy. Positive Correlation will signify that knowing the structural properties of a Network will assist us to know the prediction accuracy in advance.

Two Level Empirical Study of Logging Statements in Open Source Java Projects

Log statements present in source code provide important information to the software developers because they are useful in various software development activities. Most of the previous studies on logging analysis and prediction provide insights and results after analyzing only a few code constructs. In this paper, the authors perform an in-depth and large-scale analysis of logging code constructs at two levels. They answer nine research questions related to statistical and content analysis. Statistical analysis at file level reveals that fewer files consist of log statements but logged files have a greater complexity than that of non-logged files. Results show that a positive correlation exists between size and logging count of the logged files. Statistical analysis on catch-blocks show that try-blocks associated with logged catch-blocks have greater complexity than non-logged catch-blocks and the logging ratio of an exception type is project specific. Content-based analysis of catch-blocks reveals the presence of different topics in try-blocks associated with logged and non-logged catch-blocks.

Impact of topological properties over link prediction based on node neighbourhood: A study

Link prediction is an important task to identify the future interactions among existing non-connected members of a network, by measuring the proximity between nodes in a network. Link prediction is done on the basis of node similarity. Node neighbourhood based techniques are most popularly used for link prediction. Techniques based on Node neighbourhood can be applied on various applications like biological protein-protein interaction network, social network, information network and citation network to predict the future nodes. Every network has got certain topological properties. Topological properties play an integral role in choosing a technique for a particular network for predicting future links. This paper describes the impact of topological properties over link prediction technique. It also elaborates how presence of topological properties in a network improves the efficiency of the link prediction techniques.

NRFixer: Sentiment Based Model for Predicting the Fixability of Non-Reproducible Bugs.

Software maintenance is an essential step in software development life cycle. Nowadays, software companies spend approximately 45% of total cost in maintenance activities. Large software projects maintain bug repositories to collect, organize and resolve bug reports. Sometimes it is difficult to reproduce the reported bug with the information present in a bug report and thus this bug is marked with resolution non-reproducible (NR). When NR bugs are reconsidered, a few of them might get fixed (NR-to-fix) leaving the others with the same resolution (NR). To analyse the behaviour of developers towards NR-to-fix and NR bugs, the sentiment analysis of NR bug report textual contents has been conducted. The sentiment analysis of bug reports shows that NR bugs’ sentiments incline towards more negativity than reproducible bugs. Also, there is a noticeable opinion drift found in the sentiments of NR-to-fix bug reports. Observations driven from this analysis were an inspiration to develop a model that can judge the fixability of NR bugs. Thus a framework, NRFixer, which predicts the probability of NR bug fixation, is proposed. NRFixer was evaluated with two dimensions. The first dimension considers meta-fields of bug reports (model-1) and the other dimension additionally incorporates the sentiments (model-2) of developers for prediction. Both models were compared using various machine learning classifiers (Zero-R, Naïve Bayes, J48, random tree and random forest). The bug reports of Firefox and Eclipse projects were used to test NRFixer. In Firefox and Eclipse projects, J48 and Naïve Bayes classifiers achieve the best prediction accuracy, respectively. It was observed that the inclusion of sentiments in the prediction model shows a rise in the prediction accuracy ranging from 2 to 5% for various classifiers.

ECLogger: Cross-Project Catch-Block Logging Prediction Using Ensemble of Classifiers

Background: Software developers insert log statements in the source code to record program execution information. However, optimizing the number of log statements in the source code is challenging. Machine learning based within-project logging prediction tools, proposed in previous studies, may not be suitable for new or small software projects. For such software projects, we can use cross-project logging prediction. Aim: The aim of the study presented here is to investigate cross-project logging prediction methods and techniques. Method: The proposed method is ECLogger, which is a novel, ensemble-based, cross-project, catch-block logging prediction model. In the research We use 9 base classifiers were used and combined using ensemble techniques. The performance of ECLogger was evaluated on on three open-source Java projects: Tomcat, CloudStack and Hadoop. Results: ECLoggerBagging, ECLoggerAverageVote, and ECLoggerMajorityVote show a considerable improvement in the average Logged F-measure (LF) on 3, 5, and 4 source→target project pairs, respectively, compared to the baseline classifiers. ECLoggerAverageVote performs best and shows improvements of 3.12% (average LF) and 6.08% (average ACC – Accuracy). Conclusion: The classifier based on ensemble techniques, such as bagging, average vote, and majority vote outperforms the baseline classifier. Overall, the ECLoggerAverageVote model performs best. The results show that the CloudStack project is more generalizable than the other projects.

Web navigation prediction using Markov-based models: an experimental study

Navigation prediction on web has gained a lot of importance due to its rapid growth. Web navigation prediction WNP deals with discovering the future pages for a particular website that users/the user may visit based on previously accessed web pages. WNP can be applied effectively in many applications such as web related search, latency reduction, improving website design, location prediction and personalised systems. Markov models and its variations are widely used to represent and analyse the navigational data on web. It can also used to predict user navigation behaviour. Diverse Markov model varies in terms of state-space complexity, prediction accuracy, model accuracy, coverage and failure cases. Research is still in progress for determining the suitability of Markov models for various web applications. Therefore, this paper presents a comparison of existing Markov-based models and its merit lies in the non-trivial conclusions derived by the experiments. Comparison is done in terms of state-space complexity, failure cases, coverage, prediction accuracy and model accuracy.

Prediction of missing links in social networks: feature integration with node neighbour

Link prediction techniques are used to identify the future network structure on the basis of existing connectivity pattern of the users. Most of the existing link prediction techniques employ varied similarity indices to predict new links in network. Some techniques use common neighbours while others use common shared profile information of the user for prediction. Typically existing link prediction techniques have only focused on one of these two data modalities: common neighbours or common attributes. Both of them play equally important role in the dynamics of the network. In this paper, we propose a feature integrated node neighbour (FINN) approach, an accurate algorithm for predicting links in network. FINN integrates Jaccard coefficient and Adamic Adar to predict link between nodes using their connections and features. We have evaluated FINN by implementing it over the real-time Facebook dataset collected from SNAP repository and validated the result through area under ROC curve.

Machine Learning or Information Retrieval Techniques for Bug Triaging: Which is better?

Bugs are the inevitable part of a software system. Nowadays, large software development projects even release beta versions of their products to gather bug reports from users. The collected bug reports are then worked upon by various developers in order to resolve the defects and make the final software product more reliable. The high frequency of incoming bugs makes the bug handling a difficult and time consuming task. Bug assignment is an integral part of bug triaging that aims at the process of assigning a suitable developer for the reported bug who corrects the source code in order to resolve the bug. There are various semi and fully automated techniques to ease the task of bug assignment. This paper presents the current state of the art of various techniques used for bug report assignment. Through exhaustive research, the authors have observed that machine learning and information retrieval based bug assignment approaches are most popular in literature. A deeper investigation has shown that the trend of techniques is taking a shift from machine learning based approaches towards information retrieval based approaches. Therefore, the focus of this work is to find the reason behind the observed drift and thus a comparative analysis is conducted on the bug reports of the Mozilla, Eclipse, Gnome and Open Office projects in the Bugzilla repository. The results of the study show that the information retrieval based technique yields better efficiency in recommending the developers for bug reports.