Mining Multiple Body Measurements of Males for the Feasibility of Assigning a “Personal Code”
by
C.N. Kuruwita
Department of Statistics,University Of Colombo, Sri Lanka
Coauthors: Dr W.N. Wickremasinghe (Department of Statistics,University Of Colombo, Sri Lanka)

Due to the inherent rigid structure in the conventional size regiments in ready-made garments, a fair proportion of people find it difficult to wear them without alterations. The present study explores the possibility of assigning a “Personal Code” to an individual that could condense all his / her measurements into a single value, through an application of Statistical Data Mining. It is hoped that the new code will minimize the risk of misspecification and provide an alternative size regiment for ready-made garments. Measurements on head circumference, collar, sleeve, waist, length of leg and foot were collected from 409 Sri Lankan males in the age group 21 to 28, and are used to identify natural constellations of individuals, with the help of statistical and artificial neural networking tools. Preliminary analysis of the data is used to show that the highest correlation exists between Collar size and Waist and that Waist size accounts for much of the total variation. Based on Principal Component analysis it is shown that first three components having meaningful descriptions of shape and structure of a person. The three components account for nearly eighty percent of the total variation. Factor analysis is also used to confirm these results. Further, it is shown that 18 different clusters can be identified as a basis for a “Personal Code”, based on the Ward’s clustering technique which is shown to outsmart other techniques used. A statistical rule that outperforms the neural networking method is devised to classify a new individual to one of those clusters with over ninety percent accuracy. Further research based on more records may be helpful in finding confirmatory results.

Date received: October 4, 2002