SajjadAli54 / cataract-classification

Cataract classification

Methodology

Data Visualization

5 techniques

• BGR : Blue Green Red
• RGB : Red Green Blue
• Graye Level
• Adaptive Threshold Gray Level
• Gray Level Co occurrence Matrix (GLCM) features 'dissimilarity', 'contrast', 'homogeneity', 'energy', 'ASM', 'correlation'

Data Augmentation

Augmented by Gaussian Filter
30% data of each class is augmented and then saved.
30% increase in data (600 to 750+)

Feature Extraction using GLCM

'dissimilarity', 'contrast', 'homogeneity', 'energy', 'ASM', 'correlation'

Models (Five models)

SVM, KNN, Logistic Regression, Random Forest, Naive Bayes

Selection models technique

Learning curve
Validation Curve

Both use Stratified KFold cross validation by default

Metrics

Accuracy, Precision, Recall, F1 Score, and AUC ROC score

Conclusion

KNN performs better most of the time. It gave a better accuracy of almost 64% with auc roc score of 
0.90 as  compare to other five models. Analyzed overfitting, underfitting, variance, and bias too.