Clustering and Disaster Message Classification
A three-person university project (QUT IFN580) covering unsupervised learning and modern deep learning, with teammates Luke and Daniel. The group work clustered a bike-share dataset with K-means and K-Prototypes, and fine-tuned BERT and BERTweet with Hugging Face Transformers to sort disaster messages into related and not related. The fine-tuned models beat a TF-IDF logistic regression baseline. My individual parts were extra clustering analysis, a visualisation of the BERT classifier's attention, and fine-tuning a Qwen2.5 language model for question answering on a small SQuAD subset.
UniversityMachine LearningAIPython