Tweet Analysis to check if the user is stressed

Create the environment

Install python on your system
Install Anaconda or mini Conda in your system
Create a virtual environment:
    1. conda create -p venv python==3.9
    2. conda activate venv/ 

Project Setup:

pip install -r requirements.txt

Directory Structure:

|- assets                       --- This contains the pickle file
    |- hashtag_pipeline.pkl
    |- text_pipeline.pkl
    |- img                      --- Contains image of the wordcloud for stressed and non stressed
        |- output_notstressed.png
        |- output_stressed.png
|- data
    |- twitter_stress.xlsx      --- Contains the dataset used in this project [/kaggle/input/stress-detection-from-social-media-articles/Twitter_Full.csv]
|- .gitignore
|- model.ipynb                  --- Contains the jupyternotebook where basic EDA, feature engineering and model training was done
|- README.md
|- requirements.txt             --- Libraries Required to run this project

Algorithms used

Our problem statement here was the classification problem with dataset to identify from a tweet, whether a person is stressed or not. To achieve this, initially I used CountVector to vectorize the text data as I had planned to use Logistic Regression for this.

Post the vectorization I used AdaBoostClassifier with LogisticRegression as the classifier. I decided to go with the AdaBoostClassifier along with LogisticRegression as the data we were trying to train and test was a text data. And it might have not performed well with the LogisticRegression. Plus using an Ensemble learning method like AdaBoostClassifier has multiple benefits over using the Classification model directly like - Better performance by combining multiple weak learners, handling complex dataset, robustness to overfitting as well as better generalization.

Output