Published Paper Details:

Modern software systems, especially in e-commerce and microservice architectures, face challenges in predicting the impact of code changes due to complex interdependencies. Even minor modifications can cause cascading failures, affecting performance and reliability. Traditional approaches such as manual analysis and static dependency mapping often fail to capture dynamic behaviors and hidden dependencies. This paper proposes a lightweight, data-driven framework for offline change impact prediction that preserves data privacy. The system integrates Git-based change extraction, Abstract Syntax Tree (AST) analysis, dependency graph construction, and machine learning models (Random Forest and Logistic Regression) trained on historical data to estimate risk with probabilistic confidence scores. It also incorporates pattern mining to detect recurring failure trends. The framework consists of five modules: Change Input, Dependency Mapping, History & Pattern Learning, Risk Prediction, and Risk Advisory. It identifies affected components, classifies risk levels (Low, Medium, High), and provides actionable mitigation guidance. Experimental results show that the system delivers accurate risk predictions and practical decision support while operating on standard hardware without cloud dependencies. Unlike existing solutions, it ensures data sovereignty and is suitable for resource-constrained environments, addressing key gaps such as probabilistic prediction, project-specific learning, runtime awareness, and lightweight deployment.