Abstract The construction phase effort prediction is needed for assigning resources to teams of practitioners destined specifically to this phase of the software development life cycle (SDLC). Construction effort (CE) has been reported between 27.5% and 58% of the total SDLC effort causing the uncertainty of taking these percentages as reference. A support vector regression (SVR) training involves quadratic programming problems that can analytically be solved using a sequential minimal optimization (SMO) algorithm. Moreover, a Pearson VII (PUK) kernel is useful to replace a set of kernel functions commonly used by a SVR. The objective of this study is to apply the SMO with the PUK to train SVR for predicting CE. The SVR model trained with the SMO algorithm having as kernel to the PUK (SVR‐SMO‐PUK) prediction accuracy was statistically compared to those accuracies obtained from statistical regression (SR), neural network (NN), and two types of SVR. Seven international public data sets of software projects were used. Results showed that the SVR‐SMO‐PUK was better than the SR in five data sets and better than NN in two of these five data sets. It was equal than SR and NN in the remaining two data sets. It was equal than ε‐SVR and ʋ‐SVR in the seven data sets. Thus, the SVR‐SMO‐PUK is useful to software managers to predict CE.

The maintenance of software‐intensive systems (SISs) must be undertaken to correct faults, improve the design, implement enhancements, adapt programmes such that different hardware, software, system features, and telecommunications facilities can be used, as well as to migrate legacy software. A lack of planning has been identified as one explanation for late and over budget software projects. An activity of planning is effort prediction. The goal of this study is to propose the application of a stochastic gradient boosting (SGB) model for predicting the SIS maintenance effort. We compare the SGB prediction accuracy with those obtained with statistical regression, neural network, support vector regression, decision trees, and association rules. We trained and tested the models with five SIS data sets selected from the International Software Benchmarking Standards Group Release 11. The SGB prediction accuracy was statistically better than the mentioned five models in the two larger data sets. We can conclude that a SGB can be applied to predict the maintenance effort of SISs coded in languages of the third generation and developed on either mainframes or multi‐platform. The predicted effort corresponds to the aggregate of efforts obtained from the project team, project management, and project administration.

It is well recognized that the project productivity is a key driver in estimating software project effort from Use Case Point size metric at early software development stages. Although, there are few proposed models for predicting productivity, there is no consistent conclusion regarding which model is the superior. Therefore, instead of building a new productivity prediction model, this paper presents a new ensemble construction mechanism applied for software project productivity prediction. Ensemble is an effective technique when performance of base models is poor. We proposed a weighted mean method to aggregate predicted productivities based on average of errors produced by training model. The obtained results show that the using ensemble is a good alternative approach when accuracies of base models are not consistently accurate over different datasets, and when models behave diversely.

An important product measure to determine the effectiveness of software processes is the defect density (DD). In this study, we propose the application of support vector regression (SVR) to predict the DD of new software projects obtained from the International Software Benchmarking Standards Group (ISBSG) Release 2018 data set. Two types of SVR (i.e., ε-SVR and υ-SVR) were applied to train and test these projects. Each SVR used four types of kernels. The prediction accuracy of each SVR was compared to that of a statistical regression (i.e., a simple linear regression, SLR). Statistical significance test showed that υ-SVR with polynomial kernel was better than that of SLR when new software projects were developed on mainframes and coded in programming languages of third generation

Software engineering (SE) has been defined as the application of a systematic, disciplined, quantifiable approach to the development, operation, and maintenance of software. Enhancement is a type of software maintenance. SE involves software planning (SP), and SP includes prediction. In this study, we propose the application of two types of support vector regression (SVR) termed ε-SVR and ν-SVR to predict the duration of the software enhancement. A SVR is a type of support vector machine, which is a machine learning technique. Two data sets of software projects were used for training and testing the ε-SVR and ν-SVR. The prediction accuracy of the SVRs was compared to that of a statistical regression. Based on statistical tests, results showed that a ε-SVR with linear kernel was statistically better than that of a statistical regression model when software projects were enhanced on Mid Range platform and coded in programming languages of third generation.

Once a software project has been developed and delivered, any modification to it corresponds to maintenance. Software maintenance (SM) involves modifications to keep a software project usable in a changed or a changing environment, reactive modifications to correct discovered faults, and modifications to improve performance or maintainability. Since the duration of SM should be predicted, in this study, after a statistical analysis of projects maintained on several platforms and programming languages generations, data sets were selected for training and testing multilayer feedforward neural networks (i.e., multilayer perceptron, MLP). These data sets were obtained from the International Software Benchmarking Standards Group. Results based on Wilcoxon statistical tests show that prediction accuracy with the MLP is statistically better than that with the statistical regression models when software projects were maintained on (1) Mid Range platform and coded in programming languages of third generation, and (2) Multi platform and coded in programming languages of fourth generation.

Context: Productivity management of software developers is a challenge in Information and Communication Technology. Predictions of productivity can be useful to determine corrective actions and to assist managers in evaluating improvement alternatives. Productivity prediction models have been based on statistical regressions, statistical time series, fuzzy logic, and machine learning. Goal: To propose a machine learning model termed general regression neural network (GRNN) for predicting the productivity of software practitioners. Hypothesis: Prediction accuracy of a GRNN is better than a statistical regression model when these two models are applied for predicting productivity of software practitioners who have individually developed their software projects. Method: A sample obtained from 396 software projects developed between the years 2005 and 2011 by 99 practitioners was used for training the models, whereas a sample of 60 projects developed by 15 practitioners in the first months of 2012 was used for testing the models. All projects were developed based upon a disciplined development process within a controlled environment. The accuracy of the GRNN was compared against that of a multiple regression model (MLR). The criteria for evaluating the accuracy of these two models were the Magnitude of Error Relative to the estimate and a t-paired statistical test. Results: Prediction accuracy of an GRNN was statistically better than that of an MLR model at the 99% confidence level. Conclusion: An GRNN could be applied for predicting the productivity of practitioners when New and Changed lines of code, reused code, and programming language experience of practitioners are used as independent variables.

Context: In the software engineering field, only 20 percent of software projects finish on time relative to their original plan. A software project can be classified as a new development, an enhanced development or a re-development. Goal: To propose a feed forward neural network (FFNN) for predicting the duration of new software development projects. Hypothesis: The accuracy of duration prediction for an FFNN is statistically better than the accuracy obtained from a statistical regression (SR) when an adjusted function points (AFPs) value, obtained from new software development projects, is used as the independent variable. Method: A sample obtained from the International Software Benchmarking Standards Group (ISBSG) Release 11 corresponding to new development projects was used. The accuracy of the FFNN was compared against that of an SR model. The criteria for evaluating the accuracy of these two models were the Mean Magnitude of Relative Error (MMRE) and an ANOVA statistical test. Results: Prediction accuracy of an FFNN was statistically better than that of an SR model at the 90% confidence level. Conclusion: An FFNN could be applied for predicting the duration of new software development projects when AFPs were used as independent variable.

Expert-based effort prediction in software projects can be taught, beginning with the practices learned in an academic environment in courses designed to encourage them. However, the length of such courses is a major concern for both industry and academia. Industry has to work without its employees while they are taking such a course, and academic institutions find it hard to fit the course into an already tight schedule. In this research, the set of Personal Software Process (PSP) practices is reordered and the practices are distributed among fewer assignments, in an attempt to address these concerns. This study involved 148 practitioners taking graduate courses who developed 1,036 software course assignments. The hypothesis on which it is based is the following: When the activities in the original PSP set are reordered into fewer assignments, the result is expert-based effort prediction that is statistically significantly better.

The software project effort estimation is an important aspect of software engineering practices. The improvement in accuracy of estimations is a topic that still remains as one of the greatest challenges of software engineering and computer science in general. In this work, the effort estimation for shortscale software projects, developed in academic setting, is modeled by two techniques: statistical regression and neural network. Two groups of software projects were made. One group of projects was used to calculate linear regression parameters and to train a neural network. The two models were then compared on both groups, the one used for their calculation and the other that was not used before. The accuracy of estimates was measured by using the magnitude of error relative to the estimate (MER) for each project and its mean MMER over each group of projects. The hypothesis accepted in this paper suggested that a feed forward neural network could be used for predicting short-scale software projects.

The identification problem considered in this work, consists in compute an Interpreted Petri Net (IPN) model, in proportion as new output signals of the system are observed. The identification problem becomes complex when the complete state of the system cannot be fully measured. The state information that is not observed is inferred during the identification process allowing the computed model represents the observed system behavior. As the system evolves new information is revealed and the wrong dependencies are eliminated in order to update the computed model. Given this problem, in this paper are presented the needed algorithms to identify a class of Petri Nets (PN) known as state machines.