Trial of C4.5/ID3 in predicting the presence of rectosigmoid polyps: a preliminary report in National Taiwan University Hospital

In this study, the C4.5/ID3 learning method is applied to routine physical checkup data to determine the predictive value for colorectal polyps in the recosigmoid region in a general population, comparing data with sigmoidoscopic findings. As a first step, C4.5 was applied to 2746 patients to construct an antecedent-consequence rule for polyp prediction. Following this, the developed antecedent consequence rule for polyp prediction was used on another testing set of 2384 patients. Comparison with the findings on the sigmoidoscopy revealed that C4.5/ID3 failed to identify any significant risk factor for rectosigmoid polyps. Thus, it is concluded that computer simulation study attempting to combine putative risk factors in the prediction of colorectal polyps in general populations is immature.