Effects of Mahalanobis Distance and Prior Probabilities: On the Performance of the Linear and Quadratic Discriminant Functions: A Monte Carlo Approach - Softcover

Adebanji, Atinuke; Nokoe, Sagary

Softcover

ISBN 10: 3846524166 ISBN 13: 9783846524169

Verlag: LAP LAMBERT Academic Publishing, 2011

Inhaltsangabe

In Discriminant analysis, one requires a relatively large training sample to construct a discriminant function and evaluate its performance. This is often unattainable in practice and so numerous Monte Carlo studies have been undertaken in an attempt to shed light on the asymptotic properties of classification functions. This empirical study examines the asymptotic performance of normal-based Linear and Quadratic discriminant functions for observations from two multivariate normal populations with different prior probabilities and varying between group distances. The sensitivity of these functions to increase in sample size, prior probability and Mahalanobis distance is investigated. Four sample size ratios of group1 to group2 (n1:n2) considered are: (1:1), (1:2), (1:3), and (1:4). Sample sizes ranging from 25 to 3,000 were considered for values 1 to 7 of group centroid separator. Multivariate normal observations were generated for two P-variate populations with p=4, p=6, and p=8 (LDF(4), LDF(6) , LDF(8), QDF(4), QDF(6) and QDF(8)) models. The respective sample size-ratio combinations were repeated for each level of with 100 replications for each scenario.

Über diesen Titel

Über die Autorin bzw. den Autor

Atinuke Adebanji (PhD University of Ibadan, 2006) is a senior lecturer with the Department of Mathematics, Kwame Nkrumah University of Science and Technology, Kumasi, Ghana (2009). She was a lecturer of Statistics at the Univ of Agric, Abeokuta, Nigeria (2003-2009). Her postgraduate theses were both on Discriminant Analysis.

Details zur Ausgabe

• Verlag: LAP LAMBERT Academic Publishing
• Erscheinungsdatum: 2011
• Sprache: Englisch
• ISBN 10: 3846524166
• ISBN 13: 9783846524169
• Einband: Taschenbuch
• Anzahl der Seiten: 144
• Kontakt zum Hersteller: Nicht verfügbar
• Verantwortliche Person: Wydawnictwo Bezkresy Wiedzy; customerservice@vdm-vsg.de; ; Brivibas gatve 197; Riga; 1039; Lettland