Spam SMS Filtering using Recurrent Neural Network and Long Short Term Memory

SMS is the abbreviation for Short Messaging Service which uses standard protocols for mobile devices to exchange information via short text messages. Today SMS's are an easy, inexpensive and widely accepted way to communicate rather than phone calls. Spam can be described as random unsolicited messages sent at large without any authorization from the receiver. People still deal with spammer's misusing SMS's to advertise false claims and can gain access to private information of users. Emails, social media sites, review, and even Twitter has seen spammers trying to intrude everywhere with the advent of Internet. Spam appears in many forms like comments, emails, search results and personal messages where spammers tend to gain revenues. Various Machine Learning algorithms like Neural Networks have tried to detect spam messages and normal messages or ham from SMS's. These techniques can learn high level features automatically using raw data unlike traditional ways where features are selected after analysis for classification. In this research paper, we propose a new method utilizing Recurrent Neural Network (RNN) and Long Short Term Memory (LSTM) using Keras models and Tensorflow backend to detect Spam and Ham from `SpamSMSCollection' dataset available at UCI machine learning repository. Crucial preprocessing of dataset included tokenization, TF-IDF Vectorization and removal of stopwords. Overall accuracy of 98% is achieved and shows improvement from other machine learning algorithms for spam detection.