COVID_19_chest_CT_Image_Classification

Goal:

The goal of this project is using the patients' chest CT images to predict if a patient has pneumonia caused by COVID-19 , normal or has other pneumonia . This is ternary classification analysis.

Data Source :

https://www.kaggle.com/praveengovi/coronahack-chest-xraydataset

Data Cleansing

• There are three different labels which implies three different target, but last two contain so much None and incomplete information, so I combined two columns’ information to get my target.
• This data needs to be cleaned first, because the metadata (target label) does not really match the images in the train data folder . I removed the images which are not in the metadata (target label) ( the images’ names in the train folder must match and the images in the train data folder).
• The train data contains all COVID_19 patients but there are no COVID_19 images in test data so I moved 20% of COVID_19 images from the train folder into the test data folder.

EDA

We can see the difference between COVID-19 pneumonia , normal or other pneumonia . Normal ones are very obvious, the histogram shows they are very smooth, on the other hand the histograms of COVID_19 pneumonia and other type pneumonia show more fluctuations. (code source:https://www.kaggle.com/adityam1311/covid-19-x-ray-images-eda-models)

Data Pre-processing

• Oversampling method was applied, bar chart shows that those classes are not balanced.
• Load the images and resize them
• Normalized Image data
• Reshaped image data and train and test labels(target)

Modeling

• Neural Network ( early stop, regularization )
• Random Forest ( GridSearch)

Model Evaluation

Neural Network performed the best

• Training Accuracy: 0.9781
• Testing Accuracy: 0.8173
• Train data f1_Score:0.9781282284242644
• Test data f1_Score:0.8072066074203615
•