Finding suitable data mining techniques for software development effort estimation

Abstract

An organization's new projects all go through an analysis process. The data gathered throughout the study serve as the cornerstone for important choices about complexity, resources, frameworks, timetables, costs, etc. Numerous methods have been developed throughout time to make the project analysis phase simpler, but most of them are still insufficient when it comes to the accuracy of the results. Without a precise analysis technique in place, even initiatives with clear goals might unravel in the later stages. Software project management still faces challenges in producing accurate and trustworthy estimates of software effort, particularly in the early stages of the software life cycle when the information available is more categorical than numerical. Predicting the number of person-hours, or months, required for software development is seen as a difficult task in Software Effort Estimation (SEE). Project cancellation or project failure is the outcome of overestimating or underestimating the software effort. Although useful sizing tools and methods derived from function points don't take into consideration the unique project management culture of a business. Data processing techniques are being investigated as a substitute estimation method as a result of these shortcomings in recent years. This research aims to propose a mixture method of functional sizing measurement and three data processing methods for effort estimation at the first stage of projects: Generalized Linear Models (GLM), Deep Learning Neural Networks (DLNN), and Decision Trees - Gradient Boosting Machine (GBM). These models’ estimation accuracies were contrasted so as to assess their potential value for implementation within businesses. Additionally, a combined strategy that mixes the output of the many algorithms is usually recommended so as to enhance prediction accuracy and forestall the incidence of over-fitting.