An Ensemble of Machine Learning Models for Classifying Bangla-English Code-Mixed SMS as Spam or Ham

In the ever-changing world of digital communication, the proliferation of code-mixed languages raises new challenges for content classification, particularly in the context of spam identification. SMS ( Short Message Service) is a popular form of communication, but there are security flaws in it, such as an increase in spam from online fraudsters. This research aims to develop an ensemble of machine-learning models that can classify Bangla-English code-mixed SMS as spam or ham. Our method, which uses a large dataset of code-mixed SMS, includes several classification algorithms, including Logistic Regression, Support Vector Machines, Random Forests, and Multi-Layer Perceptron Classifier (MLPC). To increase the robustness of our model, we employ feature extraction techniques such as TF-IDF for code-mixing scenarios. The ensemble technique employs a voting procedure to integrate individual model predictions, enhancing accuracy while minimizing the risk of misclassification. The experimental results reveal that our ensemble model outperforms single classifiers, with a 96.92% accuracy and high recall and precision metrics. This study enhances the science of natural language processing and has practical implications for SMS filtering systems in multilingual settings, hence boosting user experience and security in digital communications.