CORMSIS Seminar - "Machine learning technique selection – (my) lesson learnt" - Dr Libo Li (Southampton) Event

Event details

Different algorithms solve distinctive problems. It's a trial and error process to choose a specific machine learning technique in a data science task. Statistical inference procedure and metric evaluation are two common options to support the decision for model selection. It remains a challenge for research and business to find out the optimal machine learning technique. Our study investigates three dimensions: statistical evaluation framework, model design activities, and automated machine learning tool. The development in statistical evaluation framework suggests promising improvements to gain a better understanding in model performance. With the proliferation of data science competitions, organizations get closer to model design activities. This allows stakeholders to become more informed of strength and weakness of the models. Meanwhile, they also become aware of the potential risks in adopting those techniques when seeking automation of the data-driven business processes.