Evaluation of the performance of a machine learning algorithms in Swahili-English emails filtering system relative to Gmail classifier
Rashid Abdulla Omar, Aris Tjahyanto · 2018 International Conference on Information and Communications Technology (ICOIACT) · 2018
In recent years, means of communication has changed from writing the letter and sending it by ordinary post to electronic mail. The unsolicited mails is among the security threat in this new technology, with millions of spam distributed daily. The social means of fighting against spammers are there but not effective and so the automatic spam filtering algorithms has been introduced to reduce the effects that are caused by spammers. This paper compares three algorithms that are running using the machine learning (ML) tool called Waikato Environment for Knowledge Analysis (WEKA) by using English-Swahili dataset that author created and the results are then compared with Gmail results that are calculated manually. The algorithms are Naïve Bayes, Sequential Minimal Optimization (SMO) and J48. The findings show that SMO gives good result compared to other algorithms with accuracy of 93.23% followed by Naïve Bayes 88.47% and J48 87.22%. Also, all three algorithms come out with good results ahead of Gmail filter has 86.26% accuracy.