Fig 1.
Working of SVM variants and their working principles.
Linear SVM. Soft Margin SVM. Nonlinear SVM.
Fig 2.
Input space or non-linearly separable. RBF Kernel transformation. feature space linearly separable.
Fig 3.
Conceptual view of random forest (RF).
RF Ensemble Learning Concept. Random created multiple trees related to the deco through bagging and random feature selection, then aggregated their predictions.
Table 1.
Summary of recent literature on churn prediction.
Fig 4.
Research flow indicating hybrid strategy.
The First of Data Preparation consists of three steps. The Second phase of Model Development also consists of two steps.
Table 2.
Dataset characteristics of the Bank Customer Churn Prediction dataset.
Fig 5.
Class Distribution of Churn Yes(1) and Non Churn(0).
Fig 6.
Balance of Class Distribution using bar chart,box chart, violin plot and statistical analysis.
Table 3.
Hyper-parameters and tuning considerations for the SVC-SDNN model.
Fig 7.
Dataset train-test distribution.
Overall bar chart of 20/80 distribution. Cross validation folds distribution. One cross-validation iteration. Class Distribution in Train-Test Split.
Fig 8.
Comprehensive workflow of the proposed approach that highlights all the steps in this technique.
Fig 9.
ANN Component(of SVC-SDNN) Training and validation loss.
Fig 10.
Normalized SVC-SDNN Confusion matrix.
Table 5.
SVC-SDNN classification Report.
Table 4.
Comparative performance of machine learning models showing accuracy and AUC-ROC values.
Table 6.
RF-SDNN Classification Report.
Table 7.
SDNN Classification Report.
Fig 11.
RF-SDNN Normalized Confusion matrix.
Fig 12.
SDNN Normalized Confusion matrix.
Table 8.
SVM Classification Report.
Fig 13.
SVM Normalized Confusion Matrix.
Table 9.
Performance comparison of machine learning models for churn prediction in banking dataset. Values represent mean ± 95% confidence interval across 5-fold cross-validation. RF-SDNN and SVC-SDNN represent hybrid models combining traditional classifiers with deep neural networks.
Fig 14.
Random Forest Confusion Matrix.
Table 10.
Statistical significance testing results comparing traditional machine learning models with their hybrid SDNN counterparts across various performance metrics.
Table 11.
Performance comparison of base and hybrid models.