Unlocking Predictive Excellence: Key Drivers of Forecasting Accuracy in Knowledge-Based Banking Transformation

Abstract

In an era where predictive analytics shape strategic banking decisions, understanding what drives forecasting accuracy is no longer optional it is foundational. This study identifies and quantifies the critical factors influencing predictive accuracy within Knowledge-Based Transformation Models (KBTMs), providing actionable intelligence for organizational model development and decision-making.

Drawing on survey data from 310 banking professionals across Ghana's commercial sector, our factor-analytic investigation reveals that predictive accuracy is systematically influenced by four primary drivers: (1) construct-level measurement quality (factor loadings >0.70, AVE >0.60), (2) data structural properties (multicollinearity VIF <3.3, sample adequacy KMO >0.80), (3) model specification complexity (path density, latent variable interactions), and (4) cross-validation strategy (fold configuration, benchmark selection). Knowledge-intensive constructs (e.g., Knowledge Creation, Q² = 0.834) demonstrate higher sensitivity to measurement quality, while performance outcomes (e.g., Employee Performance, Q² = 0.191) are more affected by sample characteristics and validation approach.

We establish that predictive tool selection (PLS-Predict vs. CVPAT) should be contingent upon these underlying factors: PLS-Predict excels when measurement quality is high and multicollinearity is present; CVPAT offers advantages when sample constraints or overfitting risks dominate. By translating statistical determinants into strategic guidance, this study empowers banking executives to optimize predictive model development, allocate analytical resources more effectively, and enhance the reliability of knowledge-driven decision-making.