loading page

Deep Learning and Extreme Learning Machine for the Diagnosis of Alzheimer's Disease
  • +1
  • Ashutosh Mishra,
  • Priya Arora,
  • Akshay Jaiswal,
  • Bireshwar Mazumdar
Ashutosh Mishra
Thapar Institute of Engineering and Technology
Author Profile
Priya Arora
Thapar Institute of Engineering and Technology
Author Profile
Akshay Jaiswal
Author Profile
Bireshwar Mazumdar
Institute of Engineering and Rural Technology

Corresponding Author:[email protected]

Author Profile


The aim of this study to analyze the performance of some of the significant methods using machine learning techniques for diagnosis the Alzheimer’s disease (AD). Deep learning methods have been widely used in the AD diagnosis but Extreme learning machine (ELM) and kernel (ELM) methods have been hardly ever used. Before the deployment of these methods for computation, we present a short review in the section on related work. A series of three different cases consisting of classification models in deep learning is used. We show the computational results of three of its methods CNN, MLP and LSTM. Data set has been taken from ADNI and has been pre-processed using PCA and 10 cross-fold validation. The dataset is divided into three cases: case1, case2 and case3. The results are evaluated using two performance measures in terms of Accuracy and Error analysis. The ranking of computation methods are measured based on its performance matrices. It is observed that the performance of the proposed study to classify subjects as infected or fit using Alzheimer’s Disease Neuroimaging Initiative (ADNI*) dataset. The three cases are shuffling of presence or absence of Principal Component Analysis (PCA), and k-fold cross-validation in our operation carried out for the diagnosis. Then, a comparative study of accuracy and error as performance measures, obtained by these methods has been performed to select the best method for prediction of AD with maximum accuracy and minimum error and it is computed that the MLP deep learning method is having maximum accuracy of 82% with least error in the case3.