An Efficient CoRXGB Approach to Estimate Effort of Scrum Projects
Shivali Chopra, Arun Malik · International Journal of Electronics and Communication Engineering · 2025
In agile software development, Story Point Estimation (SPE) is at the core of project planning, resource planning, and project timeline management. During the last ten years, many researchers have reasonably attempted to propose methods for estimating story points or tasks in agile projects. Expert judgment, planning poker, and analogy are some traditional approaches that have been widely applied but criticized because of the inherent subjectivity, vulnerability to biases, and inability to handle intrinsic complexities in user stories. Eventually, these lead to inaccurate estimates, misaligned stakeholder expectations, and suboptimal sprint outcomes. The research direction has also shifted more in recent times to machine learning-based and deep learning-based approaches that try to present more systematic estimation models driven by the data itself. However, these also face difficulties while fully capturing the nuances involved with the multifaceted nature of user stories. This paper proposes a new hybrid model for software effort estimation entitled CoRXGB. This will help through synergistically combine CNN, RNN, and XGBoost and take the strength of all: CNN for extracting contextual and textual features, the Bi-LSTM for extracting sequential and temporal relations, and XGBoost is superior at classifications. Among the key originalities of the research approach, the most important may become the strategy for hyperparameter optimization that involves Bayesian Optimization integrated with the Learning Gain Matrix. This strategy thus systematically analyzes and optimizes performance gains from various configurations of hyperparameters and hence effectively removes inefficiencies that are associated traditionally with the tuning process. This indeed lets one make better-informed and selective adjustments in reaching high performance. Then, the resultant CoRXGB model has been applied extensively to a wide array of data sourced from different Agile projects that included user stories amounting to more than 23,000 stories. The results showed a significant improvement in the accuracy after hyperparameter tuning, with Appcelerator Studio increasing its accuracy to 90.55% from 82.47% and Aptana Studio increasing from 82.57% to 90.82%, reflecting an increase of 6.44%. In different data sets, CoRXGB outperformed traditional classifiers like Logistic Regression, Support Vector Classifier, and K-Nearest Neighbors, and also outperformed advanced models like RNN-CNN and DEEP-SE. These results underpin the efficiency of the CoRXGB model in story point estimation. It not only outperforms baselines by substantial margins in precision, recall, and F1-score but also holds immense promise to improve Agile project planning processes toward more reliable and efficient software development practices.