Machine Learning for Suspicious Behaviour Detection and Churn Prediction in Telecom Customer Call Data

Auto dealerships get tens of thousands of calls every day from people who want to buy, get repairs, sell, or find work.  For car dealerships to get the most out of their customer service agents, ensure customer happiness through great experiences, and boost sales and revenue through deeper engagement with customers, it is critical to understand the purpose of all those calls. This paper introduces a framework that utilizes deep learning (DL) to analyze customer call records and identify questionable patterns of behavior, such as customer churn.  The approach utilizes a structured machine learning (ML) pipeline to clean the data, handle missing values, eliminate duplicates and outliers, apply Min-Max scaling to normalize numerical features, and use one-hot encoding for categorical variables.  48 elements from three months' worth of call detail records make up the Telecom Customer Churn dataset.   Training data makes up 80% of the processed data, whereas testing data is 20% of the total. It developed a Convolutional Neural Network (CNN) architecture to automatically extract hierarchical feature representations and identify intricate patterns in consumer behavior.   Conventional ML models such as Logistic Regression (LR), Random Forest (RF), and Decision Tree (DT) are significantly outperformed by the CNN, which achieves 96.5% accuracy, 91.7% precision, 99.8% recall, 95.6% F1-score, and 0.972 ROC-AUC.  These outcomes validate the model's exceptional capability in both classification and generalization