Advanced SMS Spam Detection Using Integrated Feature Extraction

The proliferation of spam text messages, which, as telecommunications evolve to incredible velocities today, now becomes a serious problem both among consumers and businesses. To improve the detection performance of SMS spam, the Multi-Type Feature Extraction and Early Fusion Framework proposed in this research study could be used. Our strategy is the combination of different kinds of features-the text, lexical, and contextual elements-in order to achieve one of the most powerful classification models. The lexical features incorporate the patterns of words like weird symbols and spelling mistakes of words. Textual features include n-grams such as TF-IDF. Contextual features are based on the word embeddings for semantic analysis. The first step combines these characteristics in an integrated representation for any message, which is then input to a classifier developed through machine learning. For testing and comparing performance, this framework is tested on a large dataset composed of SMS messages. The results are that the proposed method decreases false positives and negatives drastically while reflecting higher values of recall and accuracy. This mixture of elements from different characteristics contributes considerably to better performance since it makes for fine-grained content understanding in the message. Represented in such a form, the framework proposed here can itself be a viable alternative in text-based filtering applications in SMS spam detection.