Using evolutionary learning classifiers to do MobileSpam (SMS) filtering

In recent years, we have witnessed the dramatic increase in the volume of mobile SMS (Short Messaging Service) spam. The reason is that operators - owing to fierce market competition - have introduced packages that allow their customers to send unlimited SMS in less than $1 a month. It not only degrades the service of cellular operators but also compromises security and privacy of users. In this paper, we analyze SMS spam to identify novel features that distinguishes it from benign SMS (ham). The novelty of our approach is that we intercept the SMS at the access layer of a mobile phone - in hexadecimal format - and extract two features: (1) octet bigrams, and (2) frequency distribution of octets. Later, we provide these features to a number of evolutionary and non-evolutionary classifiers to identify the best classifier for our mobile spam filtering system. We evaluate the detection rate and false alarm rate of our system - using different classifiers - on a real world dataset. The results of our experiments show that sUpervised Classifier System (UCS), by operating on the the above-mentioned features'set, achieves more than 89% detection rate and 0% false alarm rate.