A Domain-Driven Classification Model to Early Detection of Individuals Having High Risk to Develop Colorectal Cancer

The aim of this research is to improve Colorectal Cancer screening trying to control an individual lifestyle to reduce the probability of developing Colorectal Cancer, detect the disease in early stages, and then accelerate the screening of risky individuals and postpone the screening of ones with low risk. In this retrospective study information of 309 individuals including 84 patients whose diagnosis had been between years 2006 to 2013 and 225 healthy individuals were collected through phone or face to face interviews and exploring patient medical records. Popular techniques to develop classification models in clouding support vector machine, naive bayes, k-nearest neighbor, and decision tree were applied. Finally actionable models were determined according to both types of measures and based on domain-driven data mining approach. The results show that most of the developed models have acceptable evaluation results in predicting lifestyles. The developed non-technical measure clearly distinguishes the value of every false negative prediction and every true positive prediction itself. And finally, the actionable classifiers have been selected for domain practitioners. Only two of all the developed classifiers could satisfy both technical and non-technical measures. The results showed developed models must not only be evaluated by technical measures, but also be evaluated by medical domain interestingness, and also their application ability to actual problem solving should be explored.