Skip to main content
ARS Home » Research » Publications at this Location » Publication #120071
Title: THE DANGERS OF CREATING FALSE CLASSIFICATIONS DUE TO NOISE IN ELECTRONIC NOSE AND SIMILAR MULTIVARIATE ANALYSES

Author
item Goodner, Kevin
item DREHER, GLENN - UNIV. OF FLORIDA
item ROUSEFF, RUSSELL - UNIV. OF FLORIDA

Submitted to: Sensors and Actuators B: Chemical
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 5/1/2001
Publication Date: 12/1/2001
Citation: Goodner, K.L., Dreher, J.G., Rouseff, R.L. The dangers of creating false classification due to noise in electronic nose and similar multivariate analyses. Sensors and Actuators B. 2001. v. 80. p. 261-266.

Interpretive Summary: The use of multivariate statistics has grown dramatically in the last 30 years. However, there is the possibility of multivariate techniques leading a researcher to erroneous results. This report uses randomly generated data along with experimental data to demonstrate the dangers of over-fitting data which creates artificial differentiation. It is suggested that a minimum of 6-to-1 ratio, data points to variables in the model, should be used.

Technical Abstract: The use of multivariate statistics has grown dramatically in the last 30 years. However, there is the possibility of over-fitting data using multivariate techniques that could lead a researcher to erroneous results. This report uses randomly generated data along with experimental data to demonstrate the dangers of over-fitting data which creates artificial differentiation. Specifically such techniques as analysis of variance (ANOVA), principal components analysis (PCA), and discriminant function analysis (DFA) are analyzed. It is suggested that a minimum of 6-to-1 ratio, data points to variables in the model, should be used.