Which models of the past are relevant to the present? A software effort estimation approach to exploiting useful past models

Minku, Leandro L.; Yao, Xin

doi:10.1007/s10515-016-0209-7

Cited by 21 publications

(32 citation statements)

References 58 publications

(120 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The studies above did not take into account the fact that SEE is an online learning problem, where new projects need to be predicted over time, and changes suffered by the company may affect the quality of existing SEE models [33,36]. With that in mind, Dynamic Cross-company Learning (DCL) [33,36] creates an ensemble of WC and CC models and dynamically identifies which of them is currently beneficial for SEE.…”

Section: Related Workmentioning

confidence: 99%

“…With that in mind, Dynamic Cross-company Learning (DCL) [33,36] creates an ensemble of WC and CC models and dynamically identifies which of them is currently beneficial for SEE. The beneficial models are emphasised when the ensemble is asked to estimate a new WC project.…”

Section: Related Workmentioning

confidence: 99%

“…The approach managed to obtain similar performance when using only training projects from the same time period as the project being estimated, and when using a mix of training projects from the same and different time periods. However, similar to DCL [33,36], this approach still relies on a good number of projects from the same time period to be available for the process of generating the SEE model.…”

Section: Related Workmentioning

confidence: 99%

“…Therefore, many studies have been attempting to tackle such challenges in order to improve the predictive performance of SEE models and facilitate their use. In particular, several studies have investigated the use of Cross-Company (CC) projects to reduce the cost of collecting WC projects for training SEE models [17,26,36].…”

Section: Introductionmentioning

confidence: 99%

“…Therefore, several studies proposed ways to tackle heterogeneity in CC SEE [20,22,26,35,36]. In particular, the approach Dycom [35] managed to drastically reduce the number of WC projects used for training while maintaining or slightly improving predictive performance in comparison with WC models.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Clustering Dycom

Minku

Hou

2017

Proceedings of the 13th International Conference on Predictive Models and Data Analytics in Software Engineering

Self Cite

View full text Add to dashboard Cite

Background: Software Effort Estimation (SEE) can be formulated as an online learning problem, where new projects are completed over time and may become available for training. In this scenario, a Cross-Company (CC) SEE approach called Dycom can drastically reduce the number of Within-Company (WC) projects needed for training, saving the high cost of collecting such training projects. However, Dycom relies on splitting CC projects into different subsets in order to create its CC models. Such splitting can have a significant impact on Dycom's predictive performance. Aims: This paper investigates whether clustering methods can be used to help finding good CC splits for Dycom. Method: Dycom is extended to use clustering methods for creating the CC subsets. Three different clustering methods are investigated, namely Hierarchical Clustering, K-Means, and ExpectationMaximisation. Clustering Dycom is compared against the original Dycom with CC subsets of different sizes, based on four SEE databases. A baseline WC model is also included in the analysis. Results: Clustering Dycom with K-Means can potentially help to split the CC projects, managing to achieve similar or better predictive performance than Dycom. However, K-Means still requires the number of CC subsets to be pre-defined, and a poor choice can negatively affect predictive performance. EM enables Dycom to automatically set the number of CC subsets while still maintaining or improving predictive performance with respect to the baseline WC model. Clustering Dycom with Hierarchical Clustering did not offer significant advantage in terms of predictive performance. Conclusion: Clustering methods can be an effective way to automatically generate Dycom's CC subsets.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Clustering Dycom

Minku

Hou

2017

Proceedings of the 13th International Conference on Predictive Models and Data Analytics in Software Engineering

Self Cite

View full text Add to dashboard Cite

show abstract

Investigating the use of duration‐based windows and estimation by analogy for COCOMO

Nguyen

Huynh

Boehm

et al. 2019

J Software Evolu Process

View full text Add to dashboard Cite

In model-based software estimation, using the right training data is a key contributor for making accurate predictions, which is crucial for the success of software projects.This study investigates the use of duration-based windows and estimation by analogy to calibrate COCOMO and assess their estimation performance. We compare these approaches as well as the use of all available historical data using the COCOMO data set of 341 projects and NASA data set of 93 projects. The results show that timing information exists in the data sets affecting estimation accuracy. Given sufficient data for calibration, using recently completed projects within short durations generates more accurate estimates than retaining all historical data or using k-nearest neighbors based on estimation by analogy. More training data spanning a long period of time may not lead to improved estimation accuracy. This study offers evidence to support the use of projects completed within recent years for training estimation models. KEYWORDSCOCOMO, duration-based windows, estimation by analogy, k-nearest neighbors, moving windows, software estimation | INTRODUCTIONCost and effort estimation is a key activity in software project management that can affect the outcome of software projects. Inaccurate cost estimates can lead to proposal rejection, financial losses, project management problems, and overall project failure. 1-3 Thus, the software engineering research community has introduced and evaluated many approaches to making reliable cost, effort, and other related predictions. 4 The most popular approach is model-based estimation which uses some algorithms and historical data to compute estimates. 5 Effort estimation models are often built or calibrated to past projects in organizations to compute the effort estimates of new projects. Thus, the performance of such models depends much on the relevance of training data. One important question is whether legacy data of past projects is useful for training estimation models.Existing studies proposed chronology-based approaches to splitting training data and investigating this challenging question. [6][7][8] These studies assume the basis that given a project p to be estimated, the model to estimate p is built using data selected from projects completed prior to the start of p. Kitchenham et al,7 which is one of the first studies that split training data chronologically, suggested the use of 30 most recent projects instead of all historical data in an organization to train and build regression models for estimation. Song et al 9 and Minku and Yao 10 investigated estimation methods that fetch one project at a time chronologically to build models, showing that parameters of the best resulting models change over time.Lokan and Mendes 6 investigated a chronological splitting method called moving windows in which estimation models are built using windows of n most recent projects (fixed-size windows). Lokan and Mendes 11 studied the effects of using windows of all projects completed within periods immediat...

show abstract

Can Expert Opinion Improve Effort Predictions When Exploiting Cross-Company Datasets? - A Case Study in a Small/Medium Company

Ferrucci

Gravino

2019

Product-Focused Software Process Improvement

View full text Add to dashboard Cite

Which models of the past are relevant to the present? A software effort estimation approach to exploiting useful past models

Cited by 21 publications

References 58 publications

Clustering Dycom

Clustering Dycom

Investigating the use of duration‐based windows and estimation by analogy for COCOMO

Can Expert Opinion Improve Effort Predictions When Exploiting Cross-Company Datasets? - A Case Study in a Small/Medium Company

Contact Info

Product

Resources

About