Enhancing Spam Call Identification with Deep Learning: A Hybrid CDR and Audio-Based Framework

The Spam Call Prediction System is developed to effectively identify and block fraudulent and unwanted calls, addressing the increasing prevalence of phone scams that threaten user security and privacy. It introduces a machine learning-based approach that leverages an ensemble of Random Forest Classifier and Gradient Boosting Classifier to enhance the accuracy of spam call prediction. Random Forest Classifier is utilized for its robustness in handling diverse datasets and its ability to minimize overfitting, while Gradient Boosting refines predictions through iterative learning from errors, ultimately improving model performance. By integrating the strengths of both algorithms, the ensemble method developed using two approaches Voting and Stacking classifiers achieves a superior balance between prediction speed and accuracy compared to using either algorithm alone. The model developed hence analyses various features of Call Data Record(CDR) as well as audio files dataset such as incoming calls, including caller behavior patterns, call history, and associated metadata, to classify calls as legitimate or spam.