Hybrid decision tree and logistic regression classifier for email spam detection

Email spam is an increasing problem because it disrupting and time consuming for user, since the easy and cheap of sending email. Email Spam filtering can be done with a binary classification with machine learning as classifier. To date, email spam detection still challenging since the email spam still happens a lot and the detection still need improvement. Decision Tree (DT) is one of famous classifier since DT able to handle nominal and numerical attributes and increasing the efficiency of computing. However, DT has a weakness in over-sensitivity to the training set and the noise data or instance that can degrade the performance. In this study, we propose hybrid combination Logistic Regression (LR) and DT for email spam detection. LR is used for reduce noisy data or instance before data feed to DT induction. Noisy data reducing is done by LR by filtering correct prediction with certain false negative threshold. In this study, Spambase dataset is used to evaluate the proposed method. From the experiment, the result shows that proposed method yield impressive and promising result with the accuracy is 91.67%. It can be concluded that LR able to improve DT performance by reducing noisy data.