Software Estimation Process Improvement using Data Segmentation

Estimating software development effort holds pivotal significance in the domains of resource allocation, budgeting, and project planning. Despite this importance, achieving precise predictions remains a complex challenge. This study aims to refine software effort assessment precision by investigating the efficacy of a segmentation technique tailored for dataset analysis. This approach involves categorizing interconnected projects into distinct segments based on their size, complexity, and domain characteristics. The inherent assumption is that tasks sharing similar attributes will coalesce within segments. Through such delineation, patterns and trends within each segment can be illuminated. This segmentation technique is anticipated to enhance the accuracy of software effort estimation through meticulous intra-segment effort computations and accounting for variations across segment boundaries. To substantiate this approach, the dataset is partitioned into subsets based on productivity values. Employing machine learning models like Linear, Support Vector Regression (SVR), k-Nearest Neighbors (k-NN), Multilayer Perceptron (MLP), and an Artificial Neural Network (ANN) with Exponential Linear Unit (ELU) activation, the study transforms nominal data into numerical form and evaluates model performance using metrics like MMRE, MAE, MdMRE, MdAE, and PRED. The segmentation technique exhibits promising outcomes in elevating software effort estimation accuracy, achieving an average MAE reduction of approximately 6.79 units compared to prior research. This technique holds potential to inform resource allocation and project scheduling decisions in the software industry, advancing estimation processes and empowering more informed strategic choices.