This project was done in my third semester at my college. I was taking a classification class and was required to complete a project using classification techniques that I had learned to find the best classifiers for my chosen dataset. Seeing how medical images are widely used in hospitals to diagnose diseases, especially chest Xray. I decided to choose a chest X-ray dataset for my final project.
In this project, I applied many classifiers including Instance-based, Bayes, Support Vector Machine, and neural networks. The purpose is to understand how each classification technique works and which one yields the best result for this image dataset.
The first step in every machine learning project is exploratory data analysis. This dataset was downloaded from Kaggle in .png format. Also, each image was read by multiple radiologists, so each one was associated with multiple labels (multiple diseases). Hence, I converted the .png format into pixel and also used the most common voting technique to have the label unique. Dimensional Reduction analysis, 95% PCA, was used to reduce the number of features from 65,536 features to 349 features.
After the EDA was done. The dataset was ready for classification. Support Vector Machine outperformed all other techniques with the highest accuracy of 90.8%. I also discovered that with classification trees and logistic regression were more computationally intensive for a high-dimension dataset.
Comments