ABSTRACT
To develop a comprehensive assessment of the major factors affecting the performance of AI-based predictive analytics models, this work uses the Fuzzy Decision-Making Trial and Evaluation Laboratory (DEMATEL) methodology for holistic evaluation. These goals we consider (not in any specific order) are data quality, feature engineering, algorithmic choice, ethics, external environmental factors, etc. Its complex interrelations and their co-influences on the performance of predictive models are analyzed. By applying a structured analysis framework in Excel, those drivers that are most important are identified systematically and the dependencies between drivers are established. Based on results obtained from causal influence (R', hereafter 'R-C') and total relationship (R' + C) measures, the findings give responsible insights into how various aspects of the predictive modeling process interact. The direct contribution of this research is worthwhile for both researchers and practitioners to enhance the decision-making and interpretability of AI models and in general to improve prediction performance in the wide spectrum of application areas, including healthcare, finance, and environmental sciences.
