Sentiment_Analysis_Mental_Health

Project Overview

• Builded a ML model to classify mental health status from text data using NLTK and XGBoost.
• Preprocessed the text, did feature engineering and apply TF-IDF as bag of words
• After tunning I saved and tested the model with a 82% accuracy

Exploratory Data Analysis

• Understanded and analized the distribution of mental health statuses in the dataset
• Visualized word clouds for different mental health statuses
• Analyzed other important features like text lenghth and average words

See the full code HERE

Data Preprocessing

• Encoded the labels and splited the data into training and test sets
• Resampled the training data to address class imbalance
• Preprocessed the text data by converting to lowercase, removing non-alphanumeric characters, and applying stemming

See the full code HERE

Model Training

• Used TF-IDF vectorization to convert the text data into numerical features
• Performed feature selection using SelectKBest and chi-square
• Trained an XGBoost classifier on the selected features
• Evaluated the model performance using classification report obtaining a 82% accuracy

See the full code HERE

Model Testing

• Loaded the trained XGBoost model, TF-IDF vectorizer, and feature selector
• Implemented a function to predict the mental health status of new text inputs
• Tested the model on sample inputs and printed the predicted labels

See the full code HERE

Conclusions

• We reached a hight accuracy (82%) despite using classical ML and TT-IDF
• We can try a lot of different approaches to improve performance like LSTM, Transformers and embbedings