Automated medical diagnosis using K-nearest neighbor classification

Health care practices involve collecting all kinds of patient data which would help the doctor correctly diagnose the condition the subject is likely suffering from. This data could be everything from the simple symptoms observed by the subject, initial diagnosis by a physician or a detailed test result from a lab. Thus far this data is only utilized for subjective analysis by a doctor who then ascertains the disease in play using his/her personal medical expertise.

We posit that there is definite potential for application of data mining routines on this rich reserve of patient data. Employing apposite data mining techniques, useful patterns and conclusions could be drawn from the raw data at our disposal. These findings could in turn be utilized in a number of productive ways like to carry out automated diagnosis, equip doctors with a better understanding of the causes and factors in play behind a particular disease. Automated diagnosis in particular could prove very useful for the determination of non-critical diseases or in cases where a doctor may not be available to carry out diagnosis such as being in a remote location. By studying different data mining techniques, we found a suitable technique for carrying out medical diagnosis using K-NN and achieved accuracy above 84%. We incorporated it in an application MedEval which will provide diagnosis, storing of patient information and provide visualization of that data.