A Python framework for exhaustive machine learning algorithms and features evaluations

Dubosson, Fabien (University of Applied Sciences and Arts Western Switzerland (HES-SO Valais-Wallis)) ; Bromuri, Stefano (University of Applied Sciences and Arts Western Switzerland (HES-SO Valais-Wallis)) ; Schumacher, Michael (University of Applied Sciences and Arts Western Switzerland (HES-SO Valais-Wallis))

Machine learning domain has grown quickly the last few years, in particular in the mobile eHealth domain. In the context of the DINAMO project, we aimed to detect hypoglycemia on Type 1 diabetes patients by using their ECG, recorded with a sport-like chest belt. In order to know if the data contain enough information for this classification task, we needed to apply and evaluate machine learning algorithms on several kinds of features. We have built a Python toolbox for this reason. It is built on top of the scikit-learn toolbox and it allows evaluating a defined set of machine learning algorithms on a defined set of features extractors, taking care of applying good machine learning techniques such as cross-validation or parameters grid-search. The resulting framework can be used as a first analysis toolbox to investigate the potential of the data. It can also be used to fine-tune parameters of machine learning algorithms or parameters of features extractors. In this paper we explain the motivation of such a framework, we present its structure and we show a case study presenting negative results that we could quickly spot using our toolbox.


Keywords:
Conference Type:
full paper
Faculty:
Economie et Services
School:
HEG-VS
Institute:
Institut Informatique de gestion
Subject(s):
Informatique
Publisher:
Crans-Montana, Suisse, 23-25 March 2016
Date:
Crans-Montana, Suisse
23-25 March 2016
2016
Pagination:
7 p.
Published in:
Proceedings of the 30th IEEE International Conference on Advanced Information Networking and Applications (AINA) 2016
DOI:
ISSN:
1550-445X
Appears in Collection:

Note: The status of this file is: restricted


 Record created 2016-09-28, last modified 2019-06-11

Fulltext:
Download fulltext
PDF

Rate this document:

Rate this document:
1
2
3
 
(Not yet reviewed)