With the current machine learning toolsets, algorithm selection and tuning can be automated. The data scientist improves matters most effectively during data exploration and feature building. Appropriate feature engineering sets models up for success. This is demonstrated by showing how features were processed in a healthcare predictive model, where a model predicted length of stay in hospital and a graduate performance model, where training data were used to predict later performance.
I studied aeronautical engineering at University of the Witwatersrand, graduating in 1994. I worked as an engineer in flight test on the Rooivalk, before working in enterprise Infotech. In 2000 I was architect for the South African Home Affairs National Identification System (HANIS), the largest system of its kind in the world at that time. I then worked on the world’s first combined criminal and civil AFIS. After some years designing enterprise software, I chose to switch careers to do predictive analytics for business optimisation. An example of my work is predicting the length of stay in hospital beds for elderly hospital patients in the UK. The model achieved an 85% accuracy level. I am most interested in NLP using type dependencies and world models.