Empirical Application of Stacked Generalization Model Based on Feature Fusion

This study analyzed the practical application of a stacked machine learning method based on feature fusion and compared its performance with traditional methods. This model utilizes advanced algorithms such as random forest, and XGBoost, taking the original single algorithm as the basic element. First, multiple basic elements are trained with the training set, and then these basic elements are allowed to predict the validation set. The prediction results are used as the input of the meta-learner for training. Finally, during the prediction, the prediction results of the basic elements are fused through the meta-learner to obtain the final detection result. This model aims to provide a real-time and automated abnormal data detection process through comprehensive data preprocessing, feature extraction and model training. Meanwhile, the model performance is improved through key data preprocessing steps, including outlier processing and the Synthetic Minority Oversampling technique (SMOTE) for resolving class imbalance. The experimental results show that this combined model is superior to the single model in terms of prediction accuracy and also performs very well in indicators such as ROC-AUC and Recall. Through SHAP analysis, it has also been confirmed that this combined model has high interpretability. This dynamic, data-driven abnormal data detection model improves the pre-discrimination ability and performance of various abnormal transactions in practical applications, and supports people to make data-supported optimal decisions on abnormal behaviors in complex conditions.