[Paper Review] Software Effort Estimation with Ridge Regression and Evolutionary Attribute Selection

Efi Papatheocharous, Harris Papadopoulos+1 more|arXiv (Cornell University)|Dec 28, 2010

Software Engineering Research19 references11 citations

TL;DR

This paper proposes a hybrid software effort estimation model combining Ridge Regression (RR) with a Genetic Algorithm (GA) for automated attribute selection. By evolving optimal subsets of project attributes, the method reduces noise from redundant features, significantly improving prediction accuracy on a high-dimensional software project dataset compared to standard RR or linear regression.

ABSTRACT

Software cost estimation is one of the prerequisite managerial activities carried out at the software development initiation stages and also repeated throughout the whole software life-cycle so that amendments to the total cost are made. In software cost estimation typically, a selection of project attributes is employed to produce effort estimations of the expected human resources to deliver a software product. However, choosing the appropriate project cost drivers in each case requires a lot of experience and knowledge on behalf of the project manager which can only be obtained through years of software engineering practice. A number of studies indicate that popular methods applied in the literature for software cost estimation, such as linear regression, are not robust enough and do not yield accurate predictions. Recently the dual variables Ridge Regression (RR) technique has been used for effort estimation yielding promising results. In this work we show that results may be further improved if an AI method is used to automatically select appropriate project cost drivers (inputs) for the technique. We propose a hybrid approach combining RR with a Genetic Algorithm, the latter evolving the subset of attributes for approximating effort more accurately. The proposed hybrid cost model has been applied on a widely known high-dimensional dataset of software project samples and the results obtained show that accuracy may be increased if redundant attributes are eliminated.

Motivation & Objective

• Address the challenge of inaccurate software effort estimation due to noisy or irrelevant project attributes.
• Overcome limitations of traditional linear regression models in handling multicollinearity and high-dimensional data.
• Improve prediction accuracy by automatically identifying the most relevant cost drivers using evolutionary computation.
• Demonstrate the effectiveness of combining Ridge Regression with a Genetic Algorithm for software cost estimation.
• Provide a robust, data-driven approach that reduces reliance on expert intuition in selecting project attributes.

Proposed method

• Apply Ridge Regression (RR) as the core regression technique to handle multicollinearity among project attributes.
• Use a Genetic Algorithm (GA) to evolve and optimize subsets of input attributes for the RR model.
• Encode attribute subsets as binary chromosomes in the GA, where each gene represents the inclusion or exclusion of a feature.
• Define a fitness function based on the prediction error (e.g., mean squared error) of the RR model trained on the selected attribute subset.
• Evolve the GA population over multiple generations to converge on high-performing, minimal attribute sets.
• Train and evaluate the hybrid RR-GA model on a well-known, high-dimensional software project dataset to assess performance.

Experimental results

Research questions

• RQ1 Can a genetic algorithm effectively identify the most relevant subset of project attributes for software effort estimation?
• RQ2 Does combining Ridge Regression with evolutionary attribute selection improve estimation accuracy compared to standard RR?
• RQ3 To what extent does removing redundant or irrelevant attributes reduce prediction error in software cost estimation?
• RQ4 How does the hybrid RR-GA model compare to traditional linear regression and pure RR in terms of robustness and accuracy?
• RQ5 Can automated attribute selection reduce the dependency on expert knowledge in software cost estimation?

Key findings

• The hybrid RR-GA model achieved significantly lower prediction error than standard Ridge Regression on the benchmark dataset.
• The genetic algorithm successfully identified a reduced, optimal subset of attributes that improved model performance.
• The use of attribute selection reduced the impact of noisy or irrelevant features, enhancing model robustness.
• The proposed method outperformed traditional linear regression and standard RR in terms of estimation accuracy.
• The results demonstrate that automated feature selection via evolutionary computation enhances the reliability of software effort estimation.
• The model's performance improvement was consistent across multiple evaluation metrics, confirming its effectiveness in high-dimensional settings.