Abstract
Maintaining software is critical, but it can be difficult to estimate quality, reliability, effort, and costs. To accurately predict these key parameters, we propose ML-PEQRM, a novel machine-learning model. A model estimates software quality and reliability based on code complexity, maintainability, and size. It also predicts maintenance costs. The proposed ML-PEQRM model utilizes code complexity, maintainability, and size as input features to estimate software quality, reliability, maintenance efforts, and costs. The dataset comprises 25 projects with 10,000 samples of code changes and maintenance activities. A 70-30 split created training and test datasets. Conventional estimation approaches have limitations including 25% average error, unreliable predictions, and resource inefficiency. Static code attributes related to complexity and prior changes increasing complexity by 10% were most informative. Integrating product and process data decreased maintenance costs by 25% and improved reliability by 20%. Novelty lies in integrating essential metrics for maintenance cost estimation and deriving new metrics using machine learning. Static code attributes and change metrics are identified as most significant features. Novel metrics further improve performance. This makes valuable contributions by developing an accurate, practical model that organizations can leverage to enhance planning and efficiency of software maintenance activities. By leveraging code complexity, maintainability, and size as inputs, the ML-PEQRM model provides a data-driven approach improving accuracy and reliability of quality, reliability, maintenance, and cost estimation to 99%. This enables optimization of maintenance costs, reduction in downtime, and predictive maintenance. It allows development of predictive models to enhance the accuracy of maintenance operations to 99%.
Keywords: Cost Estimation, MGFPA, Machine Learning, Random Forest, Stacking Classifier, Software Maintenance, XGB.