Building Bi-lingual Anti-Spam SMS Filter

Heba Adel ,Dr. Maha A. Bayati

Short Messages Service (SMS) is one of the most popular telecommunication service packages that is used permanently due to its affordability and do not need the internet service. The growth of using SMS leads to the increase of SMS spam problem. So, SMS spam filter become a goal of many organizations to deal with those spams. This work proposes a spam classifications approach using "Naïve Bayesian" (NB) bi-lingual classifier. Based on the content; body of short messages, this classifier categorize input English/Arabic (E/A) messages as being Ham (legitimate) or Spam (unsolicited). As is the tradition, each message's body is represented by as set of features. These features are to be extracted from E/A SMS provided by certain datasets. The proposed filter was exterminated to measure it's efficiency under different settings of working permeates. For English SMS dataset, a total of 5574 SMS were considered; 70% for training and 30% for testing. For a total of 15-featuers, extracted from each SMS, an accuracy of 93% was achieved. For Arabic SMS, a total of 400 SMS were considered and under the same specifications for the English SMS, an accuracy 85% was reached. Using features selection, accuracy level was raised up to 95% for English SMS and 88% for Arabic SMS.