Title:	EXPLAINING BLACK-BOXMACHINE LEARNING MODELS
Author(s):	Krzysztof Dzieciolowski
ISBN:	978-989-8533-92-0
Editors:	Ajith P. Abraham and Jörg Roth
Year:	2019
Edition:	Single
Keywords:	Machine Learning, Black-box models, Inputs, Predictions
Type:	Poster
First Page:	236
Last Page:	238
Language:	English
Cover:
Full Contents:	click to dowload
Paper Abstract:	Despite the advances of Machine Learning (ML), the models are still being considered black-boxes that are difficult to explain. In order to address the need to interpret the ML models, I suggest the novel Input Prediction Association (IPA) plots which are based on the means of inputs and predictions, calculated in the model’s prediction ranks. The resulting plots provide a model-agnostic method of visualizing associations between the inputs and predictions. The IPA plots can help detect inputs that are linearly correlated with predictions as well as help identify the pattern of their relationships. The IPA plots can also be used to establish inputs’ thresholds that correspond to high predicted probabilities for the event of interest, thus extending the interpretation of the model inputs to an individual observation level. Such an approach can be applied in Customer Relationship Management since it suggests not only who to target but also it provides insights as to why to target specific individuals.We illustrate the IPA approach with an example from the field of home equity loan risk analysis in banking.
	Go Back