IMDB Review Classification

Overview

A machine learning project focused on sentiment analysis of IMDB movie reviews using multiple approaches, including BERT, SVM, and neural networks. The project achieves up to 87% accuracy in classifying reviews as positive or negative, demonstrating the effectiveness of different natural language processing techniques.

Technical Details

• BERT-based classification with 87% accuracy
• Support Vector Machine implementation achieving 83.5% accuracy
• Neural network architecture with 79.5% validation accuracy
• Comprehensive text preprocessing pipeline using NLTK
• Dataset of 50,000 movie reviews

Implementation Approaches

BERT Model

Implemented using Hugging Face's transformers library, the BERT model achieved the best performance with:

• Precision: 0.88 for negative, 0.87 for positive reviews
• Recall: 0.86 for negative, 0.88 for positive reviews
• F1-Score: 0.87 across both classes

Traditional Machine Learning

Implemented multiple classical approaches including:

• Support Vector Machine with PCA dimensionality reduction
• Multi-layer Perceptron neural network
• Logistic Regression with L1 regularization

Technical Capabilities

Acknowledgments

Special thanks to the Aladdin Persson YouTube channel for guidance on BERT implementation, and to Kaggle for providing the IMDB dataset.