System to classify .eml files using naive bayes algorithm.
See http://www.math.kit.edu/ianm4/~ritterbusch/seite/spam/de
The system has already been trained using spam and ham mail. The data is stored in a sqlite database under /data.db
This database is being used to classify further .eml files. The calibration is done using more .eml tests files which brings the alpha value to 0.0008. This value may be adjusted accordingly.
Currently the system classified hams as follows: When testing 1510 files the system classifies 19 files as spam. This is a precision of 98.74%
and spams as follows: When testing 222 files the system classifies 202 files as spam. This is a precision of 91%
There are several command for the application using command line arguments.
The first argument always sets the file or directory depending on the command:
- reset - Resets the database configured in Database.java
- classify - Outputs the probability of the .eml file to be a spam
- test - Runs tests over a directory of .eml files and tells the precision
- trainHam - Trains a directory of .eml as ham ! Always make sure to run a reset before training the system
- trainSpam - Trains a directory of .eml as spam ! Always make sure to run a reset before training the system