Hybrid Artificial Intelligence Approach in Improving the Accuracy of Churn Predictions on Big Data

Abstract

The explosion of digital data has given birth to the era of big data, which presents great opportunities as well as significant challenges in knowledge extraction. Traditional data mining processes often face obstacles in terms of accuracy and efficiency when faced with massive data volume, variety, and speed. This study aims to propose and evaluate a hybrid model based on Artificial Intelligence (AI) to improve the performance of the data mining process on large-scale data sets. The proposed model integrates the power of Random Forest's algorithm in handling structured data and resistance to overfitting, with the ability of Neural Networks to model complex non-linear relationships. The research uses a case study on customer churn data from the e-commerce industry which contains 1.5 million records, with comprehensive data mining process stages, ranging from data preprocessing, feature engineering, to model implementation. The results of the evaluation showed that the hybrid model achieved an accuracy of 94.7% and an AUC (Area Under the Curve) value of 0.97, significantly outperforming the Random Forest (91.2% accuracy, 0.93 AUC) and Artificial Neural Network (92.5% accuracy, 0.95 AUC) models. Although hybrid models require slightly higher computational times, the substantial increase in accuracy provides a strong justification for their use in critical business scenarios. This study provides empirical evidence that the hybrid AI approach is an effective and promising strategy to address the challenges of big data analysis, particularly in critical business scenarios where predictive accuracy is a top priority.