We all know how annoying is to get a lot of Spam messages in our inbox. This project aims to make a predictor for those and get the basics of a model deploying
- Dataset downloaded from Kaggle (aprox 5k sms message).
- Some basics data exploration to get to know the data.
- Created a model with a 96% F1 Score on validation dataset.
- Deployed it using Flask and Heroku.
In this pare I've tried to find some relation between a message characteristic and its label.
This histogram makes it clear that Spam messages tend to be bigger in size when compared to not spam
Did some basic cleaning on the text (removing stop words/ lowering the text) and used CountVectorizer to transform it and train the model
At it is a classification problem and the dataset in imbalanced decided to user F1 Score as metric to calculate the model effectiveness.
That beein said, four models were tested and it`s results are shown below
- Linear Regression: 0.96
- Decision Tree: 0.92
- Random Forest: 0.93
- Naive Bayes: 0.96
After that also tried to improve its performance by oversampling the data but the results weren't any better and decided to stick with the Naive Bayes model.
Created a simple flask app and modified some pre-existing html pages to deploy the final model on Heroku.
Page link -> https://sms-detector.herokuapp.com/
