Abstract:
Predicting customer churn in a pay-as-you-go (PAYG) model is crucial to identifying
customers who are likely to stop using the service shortly. To make the most
effective financing decision, we employ a genetic algorithm (GA) based optimization
method. GA is preferred because of its convergence and adaptability when dealing
with multi-objective optimization issues including credit evaluation, portfolio
optimization, and bank lending decisions. To further improve the performance of
the model and reduce bias, we use Synthetic Minority Over-sampling Technique
(SMOTE) in the training process. This technique generates synthetic samples of the
minority class, thereby increasing the number of samples in the minority class and
making the model less likely to be biased. In addition, we use grid search to
systematically explore the hyperparameter space, training and evaluating a model
for each combination of values. This approach allows us to find the optimal
combination of parameters that lead to the best performance. Based on the literature
review, boosting algorithms have the highest prediction accuracy. Among the
boosting algorithms, Gradient Boosting classifier performs generally better in the
base model followed by Random forest (RF). RF is preferred most due to it being less
prone to over-fitting than other algorithms. GA-GBC results to accuracy of 84.28%
while GA-RF of 84.14%. Additionally, by relevant finetuning and parameter search
during training and boosting algorithms, increases performance and accurate
predictions.