Handling imbalanced classification problem for the prediction of stroke in patients.
• Developed a framework for learning healthcare stroke data, to predict stroke incidence in patients, with imbalanced class distribution via incorporating different classification algorithms and resampling strategies in Python.
• The classifiers Logistic Regression, kNN and SVM were applied to the original, under-sampled (random), over-sampled (random, SMOTE), and combination of over and under-sampled (SMOTE-Tomek) data sets.
• Improved the prediction performance by about 33% through implementation of kNN classifier and SMOTE-Tomek resampling strategy.
• An AUC of 0.9943 and an accuracy of 97.19% were achieved on the testing set.