1993 Volume 20 Issue 1 Pages 91-106
This paper reports a case study in the use of eleven methods of discrimination procedures applicable to binary data for the purpose of classifying survivors among 952 patients (test sample) who were treated for burn injuries between May 1976 to October 1977 in different hospitals in India. These techniques were initially used for predicting survival rates on the basis of information of involvement of six major body parts obtained from 582 cases admitted to a hospital in Lucknow city (training sample) during April 1970 to April 1976. The goal of this study was to compare the performances of discrete classification rules with that of commonly used Fisher's Linear discriminant function and Logistic discrimination procedures in the presence of sparse data. The results of this study indicated that LDF performed as good as any discrete procedures and also logistic discrimination rule for the purpose of minimising the overall misclassification rates. However, in the context of burn injuries problem where the aim is to identify highest proportion of survivals correctly, Logistic discrimination rule emerged as the single best rule.
