Sean P. Rogers - Introduction to Machine Learning for Text Analysis and Classification with Python

Machine learning pipeline focuses on text preprocessing, feature engineering, and model training/evaluation using Python libraries like NLTK, scikit-learn, and pandas

Dataset consisted of ~1000 labeled tweets about wildlife selfies, categorized into classes like abusive, benign, and educational interactions

Key preprocessing steps include:

• Removing stop words, punctuation, usernames
• Lemmatization for word normalization
• Emoji handling
• Text vectorization using TF-IDF

Random Forest classifier performed well for this use case with ~90% F1 score average, preferred over SVM due to better explainability

Cross-validation and confusion matrices used to evaluate model performance and reduce overfitting

Feature engineering through one-hot encoding of key terms/signals helped distinguish between classes

Temporal analysis revealed spikes in wildlife selfie activity during vacation periods (June/July, March break)

Focus on making models explainable and accessible to non-technical stakeholders rather than pursuing maximum accuracy

Important to explore data through visualization and manual review before building models

Classical ML approaches can be preferable to deep learning/LLMs when explainability and reproducibility are priorities