<![CDATA[Big data complexity demands integration of accurate machine learning (ML) with interpretable visual analytics (VA). Traditional ML models face transparency challenges, while pure VA systems are limited in multidimensional pattern recognition. This study synthesizes 15 peer-reviewed articles (2021-2025) to evaluate ML-VA integration effectiveness in data-driven business decision-making. We identify five primary visualization designs (interactive dashboards, heatmaps, bubble charts, network graphs, counterfactual visualization), three feedback mechanisms (real-time, user refinement, interactive exploration), and human-in-the-loop (HITL) implementation for algorithm transparency. Results demonstrate Model M3 (SHAP/LIME+Network Graphics) achieves ROC-AUC 0.941, F1-Score 0.921, Accuracy 0.924, and Precision 0.931—exceeding traditional baseline by 16.7% on ROC-AUC. Critical improvements occur in model transparency (+170.5%), interpretability (+215.9%), and user engagement (+118.7%), without compromising predictive accuracy. Hybrid BI implementation yields significant business impact: process efficiency +35%, cost reduction -27%, analytical accuracy +44%, data processing capacity +85%. Structured HITL mechanism ensures meaningful human input, complete audit trails, and continuous model improvement. Evaluation framework encompasses confusion matrix, multi-metrics (accuracy, precision, recall, F1, specificity, ROC-AUC), and internal-external validity. The primary contribution is the proposed Hybrid BI Architecture that synergizes automatic ML capabilities with human domain knowledge, creating a responsible AI ecosystem with robust governance, full transparency, and measurable accountability for superior organizational decision-making in the digital transformation era.]]>
Copyrights © 2025
