Training a spam classifier to differentiate between spam and ham emails. The dataset used for training is the SpamAssassin public mail corpus which consists of a seleciton of mail messages, labelled as spam or ham.

The goal is to train a classification algorithm to differentiate between spam and ham emails.

To reach the goal, several classification models are first trained on the Apache SpamAssassin dataset. After evaluating each classifier, the 3 best performing one are fine-tuned and re-evaluated again. The 'best' classifier is then saved as .pkl file.

All the steps are contained and documented in the Jupyter Notebok Spam classifier. To lighten the amount of line of codes in the notebook, all the functions needed have been collected in different scripts - the notebook will take care to call and run such functions.

The classifiers used are:

• SGDClassifier
• MLPClassifier
• DecisionTreeClassifier
• RandomForestClassifier
• AdaBoostClassfier
• KNNClassifier
• NaiveBayes
• (Linear) SVM

The above classifier are trained on the SpamAssassin public mail corpus which consists of a selection of mail messages, labelled as spam or ham.

Each classifier is evaluated using the following performance measures:

• confusion matrix
• accuracy
• precision
• recall
• f1 score

1. Check out that you have a working Python(preferably Python 3) and Jupyter Notebook installation

2. Git clone the repo and cd inside directory

3. Install requirements:

   pip install -r requirements.txt

4. Check that all the modules have been installed, and download the mail datasets :

   python startup.py

5. Use the jupyter notebook Spam classifier to download and familiarize yourself with the data, train, evaluate and fine tune the classifiers, and pick the 'best' performing ones.