Evaluation of neuropsychological tests in classification of alzheimer’s disease

Abstract

Many neuropsychological tests are available to measure cognitive declinement in a person affected by Alzheimer’s disease. To evaluate his/her current stage in dementia and also to find the disease progression, it is necessary to perform a serial assessment of tests. As a result, the huge amount of data gets collected which depends on the number of neuropsychological tests performed to examine the patient and also with the number of visits to the clinic. From the previous correlation studies, it is observed that high computational time is required to process many neuropsychological tests. Therefore, the scores obtained from these tests are subjected to attribute selection algorithms. The six different attribute selection algorithms are used to rank the attributes, but the top four ranked attributes are consistent with InfoGain and OneR attribute evaluators. So, we subject the ordered attributes from these two attribute evaluators to different classifiers with 10-fold cross-validation. The random forest classifier performed better with InfoGain and OneR attribute evaluators. Therefore, an accuracy of 99.1% and ROC area of 0.999 is obtained from the set of top four attributes, and similar results are obtained from the set of top six and seven attributes with the combination of Infogain and OneR with BayesNet classifier.