A Cloud-Based Random Forest Framework Optimized with Bat Algorithm for E-Commerce Transaction Prediction

With the continued growth of ecommerce platforms, a vast amount of transaction data is produced that relies on predictive analytics systems to be scalable, accurate and speedy. But the traditional Machine Learning models suffer a problem with Tuning hyperparameters, it requires high computation cost and is not very scalable in the cloud environment making it quite difficult when using it for real time production. It is observed that the current methods have unnecessarily been employed. As such, in this research, a new approach to boost the transaction and the customer satisfaction prediction using an artificial intelligence-based Cloud based Random Forest—Bat Algorithm (RF-BA) is proposed. In the end, the Bat Algorithm could effectively reduce the inaccuracy and computation and makes the task a very efficient hyperparameter optimization and feature selection procedure. A real-life e-commerce Customer Behavior dataset from Kaggle is used in the experiments for test scalability and computation time for the e-commerce dataset in the cloud computing environment. RF-BA obtains the 0.91 for accuracy, 0.89 for precision, 0.90 for recall, 0.895 for F1-score and 0.92 for ROC-AUC which all beats normal RF (0.80 for accuracy and 0.74 for F1-score) and Grid search optimization on RF (0.839 for accuracy and 0.75 for F1-score) in the same experiment setup on the same dataset and same train/test set splitting. The analysis of scalability shows that there is a near-linear increment in processing time from 5 to 33 s for 10,000 to 100,000 transactions, whereas computation time is brought down to 35 s. Results like these demonstrate the framework’s ability to manage enormous scale real-time ecommerce analytics.