Scientiﬁc Methodology and Architecture Design
Alzheimer’s disease typically progresses gradually in stages namely normal, mild cognitive impairment (MCI) and dementia. The project represents an end-to-end functional approach to implement machine learning techniques and develop a web application. Alzheimer’s disease methodology (an adaptation of Knowledge Discovery in Databases (KDD) process) is applied to complete the project. The project’s architecture design is two-tier. This is a continuation of this part.
Alzheimer’s Disease Progression Methodology
The analysis and modelling are done using an adapted KDD process (Fawcett, 2005). The methodology involves gathering data from ADNI, followed by data preparation which involves cleaning, handling missing values, removing duplicates and creating new features. Analysis and visualization are also done to understand the data set.
There are many columns and few rows and therefore feature selection is done to prevent the curse of dimensionality. The data set with selected features is divided into training and test set. Machine learning algorithms are applied to training data set to develop models and tested on the test data set to ascertain the generalizability. The patterns are identiﬁed to represent knowledge and the relationship between features. The models are evaluated using diﬀerent metrics e.g., normalized confusion matrix.
Architecture Design Speciﬁcation
The architecture design for the project is two-tier and consists of:
1. Client tier. Data Visualization and explanation of the output of the machine learning models are presented using Jupyter notebooks and the predictions from the model are displayed using a web-based application. The development involves the creation of the architecture and framework, deﬁning the input features that feed into a model to return the prediction values. The application is published with all the functionality to a web service to enable a user to access it. The interface is aimed to be simple and easy to use.
2. Business logic tier. It involves loading the data from diﬀerent ﬁles, feature extraction, transformation and selection to train the models and comparing these models to obtain the most eﬃcient model that predicts the diﬀerent stages of Alzheimer’s disease.
Alzheimer’s disease methodology and architecture design are enough to successfully deliver the project. The following section includes implementation, evaluation and results for classiﬁcation models using selected features, multiple machine learning techniques and metrics. A model is also trained for developing the web-based application and answer the research questions.
The project continues here.