Hybrid Churn Prediction Model Using SMOTE SVM Resampling and Genetic Algorithm Optimized ELM

Xuan Chen

Abstract


Consumer churn prediction and early warning have emerged as a major area of study in enterprise customer relationship management due to the advancement of big data technologies and intelligent decision-making systems. To improve the accuracy and stability of the churn prediction model, a novel resampling algorithm based on support vector machines and the synthetic minority class oversampling technique was designed. This technique alleviates the category imbalance problem by extracting support vector boundary samples and using the synthetic minority class oversampling technique to interpolate new samples in their neighborhoods. Next, a genetic algorithm is introduced to optimize the input weights and hidden layer bias of the extreme learning machine globally. Then, a classifier is constructed to improve the model's generalization performance and convergence stability. Finally, the resampling algorithm is integrated with the classifier to construct a complete consumer churn prediction model. The suggested churn prediction model had the highest mean average precision and F1 value of 98.5% and 0.98 on the training set, according to the performance test results. The lowest mean square error was 0.025 and 0.028 for the training set and the test set, respectively. The results of practical application tests indicated that the prediction accuracy of the proposed prediction model in five typical datasets was up to 93.17%, and the shortest average prediction time was only 0.83 seconds. In summary, the suggested methodology can successfully raise enterprise churn customer identification's accuracy and real-time performance. It provides powerful support for intelligent customer management and precise marketing decision-making in practical application scenarios.


Full Text:

PDF


DOI: https://doi.org/10.31449/inf.v49i34.9232

Creative Commons License
This work is licensed under a Creative Commons Attribution 3.0 License.