Empirical Software Risk Calculation with Software Risk Factors

This paper presents a practical, data-centric method for estimating software project risks by integrating established machine learning models with a more comprehensive view of risk—including technical, organizational, business, and personal dimensions. Unlike previous approaches that often centered solely on technical concerns or relied on qualitative judgments, this research utilizes survey responses from 723 IT professionals in Nepal to quantitatively assess multiple risk factors. Among the three regression techniques evaluated - Linear Regression, Support Vector Machines, and Decision Trees - the Linear Regression model delivered the best predictive accuracy, achieving an R² value of 0.9357. A correlation analysis also uncovered significant interdependencies among the different risk categories, reinforcing the value of a well-rounded assessment strategy. To showcase its practical use, the model was embedded in a Django-based web application. This work offers a more inclusive and validated approach to risk prediction, tailored to the context of emerging software markets, and addresses key shortcomings in earlier models related to validation, scope, and real-world applicability.