Early Diagnosis of Alzheimer Disease Using Instance-Based Learning Techniques

Alzheimer's impairs thinking and memory ability while adversely affecting the quality of life. Machine learning and computer-aided diagnosis have gained increasing attention in the medical field especially for early Alzheimer's disease diagnosis. Several techniques achieved promising prediction accuracies, however they were evaluated on pathologically unproven data sets from diverse imaging modalities making it hard to make a rational comparison among them. Moreover, many other factors such as pre-processing, important attributes for feature selection, class imbalance, missing values in the data and the imaging quality, distinctively affect the assessment of the prediction accuracy. To overcome these limitations, our proposed model is based on data preprocessing and important features selection to eliminate the class imbalance and redundant attributes while retaining most relevant features. In particular, incremental classifiers were used instead of traditional batch learner, so to enable early diagnosis without waiting for the full training dataset to be available. Incremental classifiers are suitable for real-time diagnostic system. The comparison of classification accuracies suggest that the preprocessed data can yield higher prediction accuracies as compared to pathologically unproven raw data.