COMP5318: Machine Learning and Data Mining

Learning Outcomes

• Algorithm principles & scope: Understand core concepts, strengths/weaknesses, and applicability of ML methods for **classification, regression, clustering, and reinforcement learning tasks.
• Hands-on design & evaluation: Gain practical experience **designing, implementing, and evaluating** ML algorithms with appropriate metrics, validation, and baselines.
• Software proficiency: Use modern ML libraries and workflows (e.g., scikit-learn pipelines, model selection, visual analytics) to build reproducible analyses.
• Communication: Present and interpret data, methods, and results clearly in written and verbal formats for technical and non-technical audiences.

Takeaways

This course helped me build a solid end-to-end understanding of machine learning as a practical problem-solving workflow, not just a collection of algorithms. Through supervised and unsupervised learning topics such as classification, regression, clustering, and pattern discovery, I learned how to translate messy real-world questions into well-defined ML tasks, choose appropriate baselines, and reason about the strengths, limitations, and failure modes of different methods. A major takeaway was that performance comes from the full pipeline—data preprocessing, feature representation, model selection, hyperparameter tuning, and rigorous validation—rather than any single “best” model, and that evaluation must align with the real objective using metrics that reflect trade-offs like precision/recall, calibration, and robustness. Implementing and comparing models in Python reinforced reproducibility and experimentation discipline (e.g., consistent splits, cross-validation, and controlled comparisons), while also improving my ability to interpret results and communicate them clearly to both technical and non-technical audiences. Overall, this course strengthened my foundations for applied ML engineering and data-driven product work, and it connects naturally to modern AI practice where careful evaluation, iteration, and deployment-ready thinking matter as much as model complexity.