In our future work, we hope to further investigate the factors that may impact the number of defects at the method level in a new software version, and we call on other researchers to follow suit. To address the erroneous nature of the data applied in machine learning studies, accurate feature selection is important. First, we applied the derived optimal variables in data preprocessing as features that are correlated with the number of defects. Through these variables, the values of the defect density and the rate at which defects occur at the method level can be determined. We can deduce that the defect velocity determines the increase in the number of defects in software at the method level. Note that the terms “features”, “attributes” and “variables” are used interchangeably here as they refer to the same metrics.

Phase Detected indicates the phase in the software development lifecycle where the defect was identified. Such metric doesn't have much sense, because working software is the primary measure of progress for agile projects. Defect removal efficiency is the extent to which the development team is able to handle and remove the valid defects reported by the test team. Changes incorporated have to be monitored to understand their impact on the stability of the existing system. Changes usually induce new defects, reduce application stability, cause timelines to slip, jeopardize quality, etc. Some teams also count defect density as the number of defects per module.

Review Efficiency:

Defect density is counted per thousand lines of code also known as KLOC. Let’s consider an example to calculate the defect density in software. Defect density is a mathematical value that indicates the number of flaws found in software or other parts over the period of a development cycle. In a nutshell, it’s used to determine whether or not the software will be released.

It is essential that project members agree beforehand on the categorization (and the meaning of each categorization) so as to avoid arguments, conflicts, and unhealthy bickering later. Review Efficiency is a metric that offers insight into the review quality and testing. Learn from experts who have helped hundreds of organizations mature their testing practices and build scalable operating models.

Test Plan Coverage on Functionality:

In addition, prediction results are not presented for all datasets because some of the datasets were used as the validation set. To satisfy the objectives of the current study, we compared the results obtained by training several learning algorithms using the raw and cleaned datasets. The results obtained after preprocessing show that the errors present in the datasets can hinder the proper assessment of the average performance of the learning algorithms. As seen from a comparison of the average classification performances of the six selected classification models on the ELFF datasets before and after preprocessing, as presented in Tables ​Tables55 and ​and6.6. After preprocessing of the ELFF datasets, both the LR and KNN classifiers achieved the same CA of 98.57%.

In the above multiple regression equation, which contains two independent variables, X1 and X2, these two variables are treated as being independent of each other. In the above expression, Y represents the dependent variable and, as such, depends on X1 and X2. This indicates that X1 and X2 both contain relevant information about Y. If X1 and X2 are collinear, this situation can lead to redundancy, which will hinder the ability to generate accurate prediction outcomes. When collinear variables are applied in a prediction model, the correlation among these supposedly independent variables tends to reduce the accuracy of the model [54].

How to calculate Defect Density

Defect density also makes it easier for developers to identify components prone to defects in the future. As a result, it allows testers to focus on the right areas and give the best investment return at limited resources. The experiment was conducted on a computer with a 1.70 GHz Intel(R) Core(TM) i5-3317U CPU and 4 GB of RAM running 64-bit Windows 10. The accuracy of the learning algorithms was assessed using a statistical data management tool called Orange, version 2.7. Of course, by doing so the size of the code is fixed, so using the density as a number is quite pointless, you could actually just use "Number of bugs found by QA" as a metrics. For truly agile projects, there are only Change of Requirements, not accumulated defects that you manage.

Google Gemini, or Gemini AI, is an integrated suite of large language models (LLMs) that is... What is delta testing and how are tech leaders using it to deliver what is defect density in software testing customer insights throughout agile development? That’s why the size of the software is a very important parameter while comparing the quality of the software.

What is a Defect Based Software Testing Technique?

To ensure the perfection of software, software engineers follow the defect density formula to determine the quality of the software. This means that the predicted number of defects at the method level in the Unicore1.n version is 19 plus the standard error ε. For comparison, the actual number of defects in the Unicore1.3 project is 21. Defect density can be a useful metric for evaluating the overall quality of a product, as it allows organizations to quantify the number of defects that are discovered during the testing process.

• To ensure the perfection of software, software engineers follow the defect density formula to determine the quality of the software.
• But if you can't use metrics to measure effectiveness of testing, how do you measure it?
• As seen from a comparison of the average classification performances of the six selected classification models on the ELFF datasets before and after preprocessing, as presented in Tables ​Tables55 and ​and6.6.
• The results show correlation coefficients of 60% for the defect density, -4% for the defect introduction time, and 93% for the defect velocity.
• The current study extends these findings to ascertain their applicability at the method level.
• This is because correcting an inaccuracy at an initial point is far less expensive than correcting it later.

Second, we assigned a unique identifier to each dataset to enable us to accurately identify the defective methods in the datasets. Second, for the purpose of determining the average classifier performance, we applied a different feature selection technique. For each dataset, we determined the individual rank score of each feature, denoted by rankid, to enable us to identify the relevant features and ascertain their correlations with the target methods in the dataset, i.e., the defective methods.

Types of Test Metrics

The RF model achieved an average CA of 97.14%, while the naïve Bayes and neural network models achieved CA values of 95.71% and 94.29%, respectively. Notably, the LR and KNN models appeared to tie in terms of the recall, specificity, F-score, MCC, J-coefficient and G-mean, with average results of 93.75%, 96.26%, 92.78%, 87.50% and 96.37%, respectively. The naïve Bayes classifier achieved the best performance in terms of the AUC, with an average value of 99.39%. The AUC captures all performance-related information and allows the essential links related to classifier performance to be clearly observed [55]. The KNN classifier achieved the best results in terms of the Brier score and information score, with values of 3.66% and 39.70%, respectively.

We report an experiment conducted on the Finding Faults Using Ensemble Learners (ELFF) open-source Java projects, which contain 289,132 methods. The results show correlation coefficients of 60% for the defect density, -4% for the defect introduction time, and 93% for the defect velocity. These findings indicate that the average defect velocity shows a firm and considerable correlation with the number of defects at the method level. The proposed approach also motivates an investigation and comparison of the average performances of classifiers before and after method-level data preprocessing and of the level of entropy in the datasets. The ability to estimate the numbers of software defects at both the class and method levels in advance can greatly assist software teams in ensuring the reliability of a program throughout the SDLC and can also aid in decision-making.

What are Test Metrics?

This will give you the areas that you should focus on more accurately. This could mean making sure defects have proper affected and fix visions attached when they are reported to development. It is a little bit of an effort to categorize these defects as change related and not, but it is worth it. Taking the cumulative defect counts and test execution rates, the theoretical curve is plotted. This in comparison with the actual values will trigger an early red flag that test processes need to change if the targets are to be reached.