Artificial Intelligence Driven Intelligent Computing: Integrating Cybersecurity, IoT, Cloud Computing, Financial Analytics, and Credit Risk Assessment for Digital Transformation

Abstract

Background: Digital transformation couples cybersecurity, Internet of Things (IoT) and cloud infrastructure with data-driven financial decision-making, yet artificial intelligence (AI) in these domains is usually evaluated in isolation and under optimistic conditions.
Objective: To test, under one statistically controlled protocol, where AI adds measurable value across security, cloud–edge computing, credit risk and financial analytics.
Methods: We propose a four-layer Intelligent Computing for Digital Transformation (ICDT) framework and instantiate it on five public data sources. Eight intrusion detectors were trained on NSL-KDD and tested on its official shifted split; federated learning (FedAvg, FedProx) was simulated across ten IID or label-skewed clients; seven credit models were compared on two datasets with 30-fold repeated cross-validation; and machine-learning (ML) forecasts of the VIX and S&P 500 value-at-risk were benchmarked against HAR and GARCH-t models. Inference used bootstrap confidence intervals (CIs), McNemar, corrected resampled t, Friedman, Diebold–Mariano and Kupiec/Christoffersen tests, with calibration, SHAP explanations and age-group fairness.
Results: Intrusion detectors scoring F1 of 0.97–0.999 in cross-validation fell to 0.74–0.83 on the shifted test; novel attacks were detected 38 percentage points (pp) less often than known ones; tree ensembles kept the best ranking (XGBoost AUC 0.964) but a multilayer perceptron gave the best F1 (0.834, 95% CI [0.829, 0.839]). Federated training came within 3.5 pp of centralised F1 under label skew and far exceeded isolated clients. Stacking improved credit AUC over logistic regression by only 0.017 (95% CI [0.005, 0.028]) and 0.018 ([0.009, 0.027]), without reducing expected misclassification cost, and was the worst-calibrated model on one dataset; all models approved creditworthy young applicants 7–14 pp less often. HAR beat random forest and LightGBM for VIX forecasting (Diebold–Mariano p < 0.001), and only GARCH-t passed all 5% VaR backtests.
Conclusions: The benefit of AI complexity is domain-dependent: large for stationary high-dimensional classification, small for credit scoring, and absent for persistent financial time series. A shared assurance layer for shift-aware validation, calibration, explainability and fairness is essential for trustworthy digital transformation