SMS-Spam-Classification

This project deals with classification of SMS messages as spam or ham using Naive Bayes classifier

Datasets

There are two datasets used in this project:

1. UCI SMS Spam Dataset (in English)
2. Precog IIITD SMSAssassin Dataset (in Hindi)

This project aims to be a stepping stone towards building a universal SMS Spam Classifier regardless of the language used. To accomplish this task, I used two datasets in different languages.

Testing

Preliminary testing was done on the two datasets individually to set a baseline for comparision.

Accuracies on testing set:

UCI Dataset 98.8%

IIITD Dataset 95.2%

Later on, different combinations were tried out. Their accuracies were as follows:

UCI train + IIITD test 77.18%

IIITD train + UCI test 90.41%

Combined train + Combined 97.04%

(Combined - UCI test)train + UCI test 98.49%

(Combined - IIITD test)train + IIITD 95.8%

UCI + (Combined - UCI train) 85.99%

IIITD + (Combined - IIITD train) 90.72%