ML Playground

Medical Cost Personal dataset

Kaggle | Notebook

• Use correlation matrix to identify correlation between features/label
• Enriched data with Linear regression yields 82.88 R^2 score and 20.57% MAE
• Trivial Linear regression with cross validation yields 73.39 R^2 score and 31.55% MAE

Heart Disease Binary clasification

Kaggle | Notebook

• Analyzed, normalized, visualize the data
• Use Logistic Regression to calculate probability of Heart Disease
• Calculate confusion_matrix and metrics Recall(TPR), Precision, Accuracy, FPR, F1 score
• Plot the ROC/AUC curve
• Use F1 score to identify the threshold that maximises for Recall while balancing Precision
• Produced accuracy 80.49%, Recall 87.38%, F1: 81.82%, FPR 26.7%

Titanic Survival Prediction

Kaggle | Notebook

Logistic regression

• Trivial only including numerical columns: Accuracy: 65%, Recall(Survival): 41%, Recall(Death): 89% Pessimistic model
• Filled missing values, handled categorical-data using OneHotEncoding, dropped noise
• Cleaned data produced Accuracy: 79%, Recall(Survival): 70%, Recall(Death): 85% Much better than previous model at Recall(Survival)

Decision Tree Classifier

• Train Accuracy: 83.85, Test Accuracy: 80.45 Didn't overfit
• Accuracy: 80%
• Recall(Survival): 67%, Recall(Death): 90%
• F1(Survival): 74%, F1(Death): 84%

Random Forest Classifier

• Accuracy: 80%
• Recall(Survival): 63%, Recall(Death): 91%
• F1(Survival): 71%, F1(Death): 84%