Ensemble learning

Heart failure is a severe medical ailment that significantly impacts patients’ well-being and the healthcare system. For improved results, early detection and immediate treatment are essential. This work aims to develop and evaluate predictive models by applying sophisticated ensemble learning techniques. In order to forecast heart failure, we used a clinical dataset from Kaggle. We used the well-known ensemble techniques of bagging and random forest (RF) to create our models. With a predicted accuracy of 82.74%, the RF technique, renowned for its versatility and capacity to handle complex data linkages, fared well. The bagging technique, which employs several models and bootstrapped samples, also demonstrated a noteworthy accuracy of 83.98%. The proposed model achieved an accuracy of 90.54%. These results emphasize the value of group learning in predicting cardiac failure. The area under the ROC curve (AUC) was another metric to assess the model’s discriminative ability, and our model achieved 94% AUC. This study dramatically improves the prognostic modeling for heart failure. The findings have extensive implications for clinical practice and healthcare systems and offer a valuable tool for early detection and intervention in cases of heart failure.

Accurate and early prediction of diabetes is crucial for initiating prompt treatment and minimizing the risk of long-term health issues. This study introduces a comprehensive machine learning model aimed at improving diabetes prediction by leveraging two clinical datasets: the PIMA Indians Diabetes Dataset and the Early-Stage Diabetes Dataset. The pipeline tackles common challenges in medical data, such as missing values, class imbalance, and feature relevance, through a series of advanced preprocessing steps, including class-specific imputation, engineered feature construction, and SMOTETomek resampling. To identify the most informative predictors, a hybrid feature selection strategy is employed, integrating recursive elimination, Random Forest-based importance, and gradient boosting. Model training uses Random Forest and Gradient Boosting classifiers, which are fine-tuned and combined through weighted ensemble averaging to boost predictive performance. The resulting model achieves 93.33% accuracy on the PIMA dataset and 98.44% accuracy on the Early-Stage dataset, outperforming previously reported approaches. To enhance transparency and clinical applicability, both local (LIME) and global (SHAP) explainability methods are applied, highlighting clinically relevant features. Furthermore, probability calibration is performed to ensure that predicted risk scores align with true outcome frequencies, increasing trust in the model’s use for clinical decision support. Overall, the proposed model offers a robust, interpretable, and clinically reliable solution for early-stage diabetes prediction.

Having an accurate estimation of a customer's worth is one of the more important tasks performed by banks in this modern world, especially with the profound number of customers and the complex nature of transactions, along with the massive variance in transactions. In light of this need, we develop a multi-layer stacked ensemble model specifically designed to improve the predictive performance of banking customers in Iran. The first layer consists of 4 different learners (XGBoost, CatBoost, Random Forest, and Gradient Boosting). Each model has its learning capacity and learns from customer behavior and financial characteristics in complementary ways. The second layer consists of a LightGBM classifier, which fuses (by meta-model) the outputs of the first-layer learners into the final prediction. The second set of model hyperparameters were optimized using a Reinforcement Learning (RL)-based SWO to efficiently search for optimal hyperparameters across a high-dimensional space, which is typically not well-explored using classic optimization strategies. Utilizing a repeated 5-fold stratified cross-validation approach, we were able to achieve strong predictive accuracy: Accuracy = 89.70%; Precision = 92.84%; Recall = 92.46%; F-Score = 92.61%; ROC AUC = 0.9632; all of which surpass the single models. Our results provide evidence supporting the successful application of a multi-layer ensemble with metaheuristic hyperparameter optimization in building a viable and powerful customer valuation tool for banks.

The rapid growth of digital payment systems has heightened the need for accurate and scalable methods to detect credit card fraud. This study evaluates a range of machine learning and deep learning algorithms, including Logistic Regression, Decision Tree, Random Forest, K-Nearest Neighbors (KNN), XGBoost, Convolutional Neural Networks (CNN), Baseline MLP (Multi-Layer Perceptron), and Long Short-Term Memory (LSTM), to identify effective approaches for detecting fraudulent transactions. Based on comparative analysis, Random Forest and LSTM achieved the strongest individual performance, with accuracies exceeding 96%. Building on these findings, a stacking ensemble model was constructed by integrating Random Forest and LSTM as base learners and Logistic Regression as the meta-classifier. The framework incorporates Convolutional Autoencoder (CAE) for feature extraction and Random Undersampling (RUS) with three resampling ratios (1:1, 1:5, and 1:10) to address class imbalance. Experimental results show that the ensemble model provides improved predictive accuracy compared with individual algorithms, achieving an accuracy of 99.98%, precision of 99.86%, and recall of 99.89% under a 1:10 resampling ratio. Rather than proposing a new algorithmic architecture, this study contributes a systematic and unified evaluation of widely used ML and DL approaches and demonstrates the effectiveness of integrating CAE, RUS, and a Random Forest–LSTM stacking ensemble in enhancing fraud detection performance.