Possibilities of See5 software in forecasting of life expectancy not achieving

Results of application of See5 to epidemiological domain were presented. The aim of this paper is to find out if See5 can be used for forecasting whether life expectancy will be achieved or not. The basis for forecasting are data-attributes, predictors (anthropological data, living habits, laboratory data, blood pressure measured in years 1970/71, and the status of examinee--class (alive or dead) in 1990. Data were split at random into 10 blocks with approximately the same number of cases and the same distribution of classes. Results were given in the form of decision trees and rulesets, and tested through crossvalidation. The most interesting question--what are prediction-candidates for not achieving life expectancy estimated in the year of examination--did not give a satisfactory result: the accuracy of classification was about 70%. However, a very interesting fact is that each rule consists of at least one candidate predictor, usually considered as risk factor (e.g. high diastolic blood pressure and animal fat consumption).