Alzheimer’s Disease Stage Classification using Deep Convolutional Neural Networks on Oversampled Imbalance Data

One problem that is often faced in health data sets is a high level of imbalance. Imbalanced data can mean that the data used in machine learning has an unbalanced data distribution between different classes. One of the data is Alzheimer's disease data. The data used are Dataset containing 6400 images of brain magnetic resonance imaging that is indicated as having Alzheimer's disease or not. This study used Deep Learning Architecture to classify brains affected by Alzheimer's disease and healthy brains. We produced a trained and predictive model using Deep Convolutional Neural Networks (CNN), where the data had previously gone through the oversampling stage due to an imbalance in the data distribution. From the evaluation metric table provided in this study, the highest accuracy for detecting Alzheimer's disease was obtained by ResNetl8 with an accuracy of 60.67%. Still, the shortest training time was obtained by ShuffleNet V2 with 624.4 seconds using non-oversampled. While the results obtained by using oversampled, ShuffleNet V2 provide 60.75% accuracy with the fastest training time of 1478.6 seconds. The special CNN model that the author proposes gets an accuracy of 50% using non-oversampled and 55.27% accuracy using oversampled.