Machine Learning approach for Software Effort Estimation using Combination of Principal Component Regression and Neural Network. Issue 1 (1st August 2022)

• Record Type: Journal Article
• Authors: Priya Varshini, A G; Anitha Kumari, K; Bhalavaishnavi, V S; Kalpana Devi, C; Pranesh, S
• Is Part Of: Journal of physics. Volume 2325:Issue 1(2022)
• Publication Date: 2022-08-01
• Subjects: Machine Learning -- Software effort estimation -- regression -- Hybrid approach -- PCR and NNET; Physics -- Congresses; 530.5
• DOI: 10.1088/1742-6596/2325/1/012049
• Languages: English
• ISSNs: 1742-6588

Abstract

Abstract: For software projects that deploy vital tasks, it is difficult to estimate the effort of a project. In order to anticipate a few hours of labour effort (either time or personal) to deploy or maintain the software programme, software measurement points must be used. It is difficult to predict the behavior of an application that is engaged in software development during the initial phases of the effort. The hybrid model technique is used in this paper. It is necessary to apply Supervised Learning techniques in the Machine Learning algorithm. It is further subdivided into different types, such as linear regression, logistic regression, SVM (Support Vector Machine) algorithm, Naive Bayes algorithm, PCR(Principal Component Regression) algorithm, (Neural Network)NNET algorithm, KNN (K-Nearest Neighbour) algorithm, K-means, Random Forest algorithm, Dimensionality reduction algorithms, Gradient boosting algorithm, and Ada Boosting algorithm, to name a few examples. PCR algorithm and the nnet algorithm have been utilised for hybrid method, as shown above. Predictions from JM1/Software have been used to create this data collection. With 10886 unique instances and 18 unique traits, this is a very large amount. Metrics for evaluating this system include Mean Absolute Error (MAE), Mean Relative Error (MRE), Mean Magnitude of Relative Error (MMRE), Percentage of Predictive Accuracy (PRED), and R-squared. Compared to single model approaches based on machine learning algorithms techniques, the proposed hybrid using principal component regression and neural networks produced the best results, as demonstrated by the results.