DATA SCIENCE FRAMEWORK FOR PREDICTIVE ANALYTICS: MODEL DEVELOPMENT AND VALIDATION

The growing reliance on data-driven decision-making across sectors has positioned predictive analytics as a core application of data science. Despite significant advancements in artificial intelligence and machine learning, existing predictive analytics studies often suffer from fragmented analytical pipelines and insufficient validation, limiting model reliability and real-world applicability. This study proposes and validates a structured data science framework for predictive analytics that integrates data preparation, feature engineering, model development, and validation within a unified methodological architecture. Adopting a quantitative, modeldriven research design, the study develops multiple predictive models and evaluates their performance through systematic validation procedures. The findings demonstrate that predictive models developed within the proposed framework exhibit improved accuracy, robustness, and generalizability compared to models developed using ad hoc approaches. The results further highlight the importance of continuous validation and comparative model assessment in mitigating overfitting and enhancing predictive stability. By emphasizing methodological coherence and validation rigor, the framework supports transparent and reliable decision-support systems applicable across domains such as education, healthcare, business analytics, logistics, and sustainability. The study contributes to data science literature by offering an empirically grounded framework that bridges theoretical principles with practical implementation, providing a replicable foundation for advancing predictive analytics research and application