Electronic noses synthetic database

Published: 7 December 2020| Version 1 | DOI: 10.17632/s7c74xw673.1
Contributor:
Susana Albarracin Estrada

Description

Four work scenarios are generated artificially with the chemosensors package to address the deviations present in artificial odor systems. This research leads to a total of ten databases generated from the four scenarios, in which, noise and drift concentration present are parameterized in the generated data. Prior knowledge of these values allows analysis of results using raw data instead of the traditional method for artificial odor systems, which employs feature selection.

Files

Steps to reproduce

In total, among the four scenarios, ten synthetic databases were generated. These ten databases are used to compare the traditional method used in artificial odor systems, which consists of selected characteristics in the preprocessing stage, versus raw data from the signals, without a selection of features. In the latter, it is validated whether the rapid detection method manages to find an early portion of the signal that is not affected by drift in terms of the classifier's success rates. Several experiments were performed with different classifiers to test and validate this hypothesis.

Institutions

Universidade Federal Rural de Pernambuco

Categories

Machine Learning, Electronic Nose, Application of Sensors, Drift Analysis

License