SMS Spam Detection Airflow Training DAG

Summary

The following project demonstrates the use of apache airflow train three different models for SMS spam detection, compare their performance, and declare a winner.

The dataset used can be found here https://www.kaggle.com/datasets/uciml/sms-spam-collection-dataset

The three models trained were taken from this tutorial https://www.geeksforgeeks.org/sms-spam-detection-using-tensorflow-in-python/

Gameplan

Task List

1. Find AI project and run it - DONE
2. Automate data gathering - DONE
3. automate training model - DONE
4. automate storing the model
5. automate deploying the model

Aiflow DAG stages:

1. Download dataset - DONE
2. Train each of the three models in parallel - DONE
3. Compare the models - DONE
4. Declare the winner - DONE

Steps to run

export AIRFLOW_HOME=/home/will/Projects/AIOpsDemo
airflow standalone

Model Performance Expectations

Model Comparison Total Results:

Model Name	accuracy	precision	recall	f1-score
MultinomialNB Model	0.962332	1.000000	0.720000	0.837209
Custom-Vec-Embedding Model	0.979372	0.984733	0.860000	0.918149
Bidirectional-LSTM Model	0.984753	0.985401	0.900000	0.940767
USE-Transfer learning Model	0.982960	0.958042	0.913333	0.935154