Emotion Classification ML Pipeline

This project implements a machine learning pipeline for emotion classification from text data. It uses Python, scikit-learn, and NLTK for preprocessing, feature extraction, model training, and evaluation.

Features

• Data Loading & Persistence:
  Uses a PersistenceManager class to load and save datasets and trained models.

• Preprocessing:

  • Cleans text by lowercasing, removing digits, punctuation, special characters, and stop words (with some negations kept).
  • Applies lemmatization.
  • Handles missing values in the text column.
  • Uses a scikit-learn Pipeline for modular preprocessing.

• Feature Extraction:

  • Uses TfidfVectorizer to convert cleaned text into numerical features.

• Model Training:

  • Trains a Support Vector Classifier (SVC) with a pipeline that includes preprocessing and feature extraction.
  • Performs cross-validation to estimate model performance.
  • Uses RandomizedSearchCV for hyperparameter tuning.

• Evaluation:

  • Prints F1 score, accuracy, classification report, and confusion matrix on the test set.

• Model Saving:

  • Saves the best trained model for later use.

Usage

1. Install dependencies:

   pip install -r requirements.txt

2. Prepare your dataset:
   Ensure your data file contains at least two columns: text and emotion.

3. Run training:

   python train.py

4. Model output:
   The best model is saved using the PersistenceManager.

5. Streamlit:

    streamlit run main.py

File Structure

• train.py — Main script for training, validation, and model saving.
• preprocess.py — Preprocessing functions and pipeline components.
• persistence_manager.py — Handles loading and saving of data/models.
• requirements.txt — Python dependencies.

Requirements

• Python 3.7+
• scikit-learn
• pandas
• numpy
• nltk
• matplotlib

Notes

• The pipeline is designed to be robust to missing text values.
• All preprocessing steps are encapsulated in the pipeline for reproducibility.
• Hyperparameter search is performed for SVC’s C, gamma, and kernel parameters.

---