Abstract

Expansion of the business world and the integration of different technology has increased the use of email as a communication route among business, education system and other. This has increased the threat of randomly sent bulky email to many people at a time or even promotional emails leading to consumption of storage resources and reduced customer’s efficiency. Spam email is traditionally detected which is time consuming and require a lot of manpower. Owing to this machine learning techniques provide an easier and faster approach through the use of historical dataset. Many studies employed email content which might vary but the frequency of some specific words and characters could also be used a employed in this study. Support vector machine, K nearest neighbour (with varying value of k=2,3 and 5), Gated Recurrent Units-designed with 128 input units, 2 hidden layers of 64 and 32 units and return sequence initiated True-and Autoencoder were comparatively experimented for span email classification using the spambase dataset. This dataset was normalized and the missing values were filled. SVM and GRU showed a close average accuracy of 93% and 91% respectively using 70-30 hold-out evaluation method, while auto autoencoder gave the least performance with 59% average accuracy. This study significantly showed that GRU is also a good model with structured dataset and not with only textual, time series or streaming dataset. Future works showed also balance the dataset and experiment with other machine learning models for generalizability.

Copyright of the paper named above is hereby assigned and transferred to the Nigerian Journal of Applied Science and Innovative Technology published by College of Engineering, Bells University of Technology, Nigeria.