An Efficient Churn Prediction model using ML supervised and semi supervised Learning Techniques

Churn prediction is crucial for customer retention, prompting businesses to explore advanced machine learning techniques. This paper focuses on enhancing churn prediction accuracy through a combination of supervised and semi-supervised learning methods. State-of-the-art MLalgorithms, Random Forest, Gradient Boosting, KNN andSVM were employed within a semi-supervised learning framework. Pseudo-labeling was applied to extend the training set by predicting labels for unlabeled instances, thereby augmenting the informativeness of the dataset. Two distinct semi-supervised learning models, Self-Training, and LabelPropagation, were subsequently utilized to iteratively refine predictions.The results demonstrated the effectiveness of each model, offering a nuanced understanding of their performance metrics. For Random Forest, both the supervised and Self-Training models exhibited high accuracy (0.959) and balanced precision, recall, and F1 Score metrics. However, LabelPropagation introduced a trade-off with slightly lower accuracy (0.848) while still maintaining practical efficacy. Similarly, Gradient Boosting demonstrated consistent performance across supervised and semi-supervised settings, emphasizing the reliability of these methods.The k-Nearest Neighbors (k-NN) model showcased notable performance improvements through pseudo-labeling, with Self-Training yielding results comparable to the supervised model. However, LabelPropagation demonstrated a trade-off, suggesting potential limitations in certain scenarios. The SVM model, known for its distinctive characteristics, showed consistent accuracy across supervised and semi-supervised approaches, reflecting its stability.