Alzheimer’s Disease (AD) poses significant challenges in diagnosis due to its complex nature and the intricate changes it causes in the brain.
This page showcases two distinct machine learning projects that leverage the power of K-Nearest Neighbors (KNN) and Support Vector Machine (SVM) classifiers to diagnose Alzheimer’s Disease through the analysis of glucose metabolism patterns in the brain.
These Were the Tasks during the Machine Learning Course – MAI
Alzheimer’s Disease Diagnosis using KNN Classifier
GitHub Repo URL: https://github.com/dev-nanthan/classifier-knn-diagnosis
This project aims to develop a K-Nearest Neighbors (KNN) classifier to differentiate between stable Normal Controls (sNC) and stable Dementia of Alzheimer’s Type (sDAT) based on glucose metabolism in specific brain regions. By analyzing glucose metabolism patterns, the model predicts the likelihood of Alzheimer’s Disease in individuals.
- Data Preprocessing: Standardization of features from glucose metabolism data is performed to eliminate biases and prepare the dataset for the KNN classifier.
- KNN Classifier Implementation: Utilizes the KNN algorithm, determining the most suitable number of neighbors (k) and employing Euclidean and Manhattan distances to measure the similarity between samples.
- Performance Evaluation: The classifier’s effectiveness is gauged through error rates on training and independent test datasets, with visualization tools aiding in the assessment of the model’s robustness.
- Model Optimization: Strategies like adjusting the number of neighbors and exploring different distance metrics are employed to enhance the classifier’s predictive accuracy for AD diagnosis.
Alzheimer’s Disease Diagnosis using SVM Classifier
GitHub Repo URL: https://github.com/dev-nanthan/svm-classifier-diagnosis
This project develops a Support Vector Machine (SVM) classifier to predict Alzheimer’s Disease (AD) from high-dimensional glucose metabolism data across various regions of the cerebral cortex. Utilizing linear, polynomial, and radial basis function (RBF) kernels, the project aims to classify individuals into stable Normal Controls (sNC) and stable Dementia of Alzheimer’s Type (sDAT) groups.
- Data Preparation: Involves handling glucose metabolism measurements to form training, testing, and independent testing datasets, facilitating a comprehensive evaluation of the SVM classifier.
- SVM Classification: The SVM classifier is implemented with linear, polynomial, and RBF kernels to navigate the complex patterns of AD, with adjustments in the gamma parameter to address non-linear relationships.
- Hyperparameter Tuning: Optimal parameters for each kernel type are identified through grid search and cross-validation, focusing on the regularization parameter C, polynomial degree d, and the RBF kernel gamma γ.
- Model Evaluation: Classifier performance is assessed using metrics like accuracy, sensitivity, specificity, precision, recall, and balanced accuracy, guiding the kernel and parameter selection for optimal AD diagnosis.
Why SVM is Preferred Over KNN
- Accuracy: SVM tends to outperform KNN in high-dimensional spaces, making it more suitable for the complex patterns in AD diagnosis.
- Scalability: SVM is more scalable and computationally efficient, especially important for large datasets.
- Interpretability: Despite SVM’s complex decision boundaries, the support vectors provide insights into the classification decision.
- Parameter Tuning: While both models require hyperparameter tuning, SVM’s ability to handle overfitting through regularization makes it robust.
- Model Complexity: SVM’s kernel trick allows for modeling complex non-linear relationships without increasing the computational burden.
- Computational Resources: SVM’s prediction phase is more efficient than KNN, which must compute distances to all training instances.
Thevananthan Thevarasa 