Abstract:

Sensitivity analysis is a simple technique that is based on the amount of response (output) changes when a given input is varied though its domain and can be applied to both classification and regression Data Mining tasks. This seminar presents an improved and novel sensitivity analysis approach that was recently published in conference (IEEE CIDM 2011) and journal (Information Sciences, 2013) papers. The intention is to show how visualization techniques based on sensitivity analysis results can be used to extract human understandable knowledge from supervised learning black box data mining models, such as Neural Networks, Support Vector Machines and Random Forests. Examples of such knowledge extraction is shown for both synthetic and real-world datasets (e.g., bank direct marketing and white wine quality).