Detection of spam messages with machine learning methods

Spam messages have become a major problem in digital communication, threatening users' mental health, personal security and network resources. Traditional spam detection methods suffer from low detection rates and high false positives, highlighting the need for more effective solutions. In this thesis, various machine learning based models and hybrid approaches for spam detection are examined and the model and parameter combinations that will provide the highest accuracy rate are analyzed. The proposed model was developed using both lean and hybrid approaches, incorporating Autoencoder, Multilayer Perceptron (MLP), XGBoost, Naive Bayes, Logistic Regression, Support Vector Machines (SVM), Artificial Neural Networks (ANN), Random Forest, and Gated Recurrent Unit (GRU) models. As a result of the research and testing, we propose EGMA, an ensemble learning based hybrid model that integrates a gated recurrent unit (GRU), multilayer perceptron (MLP), Autoencoder+XGBoost and majority voting algorithm for spam detection. The model improves its performance with additional statistical features using Term Frequency-Inverse Document Frequency (TF-IDF) and CountVectorizer as text vectorization techniques. The performance of the proposed model is tested on the SMS Spam Collection, Email Spam, Enron-Spam, Super SMS and UtkMl's Twitter Spam datasets, achieving high accuracy rates ranging from 95,09% to 99,28%. The results show that the EGMA model outperforms both individual methods and other works in the literature and provides a powerful solution that contributes to the effective detection of spam messages.