Abstract: |
In medical research, being able to justify decisions is generallyas important as taking the right ones. Interpretability is thenone of the chief characteristics a learning algorithm must have,in order to be successfully applied to a medical data set. Otherimportant features are seamless treatment of different data types,and ability to cope well with missing values. XCS and decisiontrees both appear to have this desirable characteristics; wecompared them on a data set regarding Head and neck squamous cellcarcinoma (HNSCC). This kind of oral cancer already been found tobe associated with smoking and alcohol drinking habits. Howeverthe individual risk could be modified by genetic polymorphisms ofenzymes involved in the metabolism of tobacco carcinogens and inthe DNA repair mechanisms. To study this relationship, the dataset comprised demographic and life-style (age, gender, smoke andalcohol), and genetic data (the individual genotype of 11polymorphic genes), with the information on 124 HNSCC patients and231 healthy controls. Results with both algorithms are presentedand analyzed. |
