Abstract: Machine learning has been used extensively for predictive analytics in a wide range of scenarios. Recently, a number of other usages of machine learning, outside of prediction, have been developed, such as basic handling of missing data. These uses will be expanded upon in this talk, and new uses will be introduced.
The new applications of machine learning form the basis of a process to enhance the identification of which particular combinations of numerous potential features are important to an outcome and how they are related to the outcome variable. Moreover, this process can reveal which specific levels of one factor are associated with different types of relationships of another feature with the outcome variable. Such pinpointed identifications will be illustrated and shown to enable an in-depth focus on key features. It will be demonstrated how such a process led to discovery of unexpected associations in healthcare data, and to potential generalizations to additional medical conditions.
The new applications of machine learning also enable efficient modeling of multiple outcome variables, given numerous potential features. It will be demonstrated how improved values in one outcome variable are related to changes in another outcome variable in healthcare data, so as to provide a more complete picture of the complex interactions in a healthcare setting.
It will be shown how these examples of improved knowledge discovery are attained through the application of machine learning techniques to achieve new functionalities. The application of new machine learning methods to various steps of a data science workflow will be presented. It will be shown how such new applications can enhance the data exploration, data cleansing, feature selection, and model evaluation steps.
Bio: Dr. Linda M. Zeger has contributed a variety of performance improvement methods and analytics techniques to a range of disciplines including health care, education, and information dissemination. Dr. Zeger is the founder and principal consultant of Auroral LLC and is a visiting research professor at Florida Polytechnic University. In these roles, as well as through positions she has held with MIT Lincoln Laboratory, Lucent Technologies Bell Laboratories, Princeton University, and Educational Testing Services, she has developed novel methods to handle dirty data to increase data utility, and has invented techniques to increase the speed and integrity of data transmission through communication and sensor networks.
Dr. Zeger holds a Ph.D. in physics from Harvard University and an A.B. in physics from Princeton University. She is the holder of a number of patents and is the author of numerous published papers.