Bilkent University
Department of Computer Engineering


Learning a Model to Suggest Best Feature Values for the Desired Class Label


Gizem Mısırlı
MSc Student
Computer Engineering Department
Bilkent University

In medicine, deciding which kind of treatment is suitable for the patient is the most challenging question for doctors. This problem can appear in many different areas in medicine. In order to overcome this challenge, we decided to develop a suggestion system. For our suggestion system, we chose our dataset from the patients who have infertility problem. For the problem of infertility there are many ongoing treatment techniques. In Vitro Fertilization (IVF) is the major treatment for infertility among the assisted reproductive technologies. This treatment includes many different types of drugs including hormones. It is an expensive treatment and if a treatment procedure fails, the couple has to wait several months in order to try again. However, the most challenging part for both doctors and candidates is what the chance of success is. For the doctor’s side, after deciding for treatment, they have to decide which treatment protocol is the best choice for the couple. In the next step, the other question they try to answer is the dosage schema for the selected drugs in the selected protocol. Complete at this time, our research gives a direction to this challenging process. The goal of this research is to learn a model that can be used to suggest best values for selected features in a way that the chance of achieving the desired result will be maximized. Therefore, we aimed to suggest the most proper value for the selected feature especially the treatment protocol and the dosage of the drugs for our problem because if the suggested feature is the most valuable one for the patient, than the chance of achieving the desired result will be maximized. If there exists a data about the previous patients that include clinical parameters, applied treatment protocols, dosages of drugs and the results of the treatment, decision support tools can be developed for doctors based on machine learning techniques. By the help of them, while deciding the treatment protocol and the dosages of the drugs doctors can be more self-confident and the chance of acquiring positive result can be increased. In order to provide this, first of all we use a baseline solution that is called Nearest Successful Neighbor based Suggestion (NSNS). After that we try to develop other machine learning techniques like Decision Tree and Naive Bayes. However, our main goal is to implement a suggestion system that is more accurate end efficient than the well-known techniques.

Keywords: machine learning in medicine,data analysis, learn to suggest, suggestion systems


DATE: 16 April, 2012, Monday @ 16:40