Classifying Disease Mutation data

Created machine learning models in R and Python to classify curves of disease data as mutation or no mutation.

The data

7 variables for learning and response variable True (is mutation) and False (is not).

R workflow

R results

All models show promising mean accuracy over 90% Random Forest performs the best with mean accuracy of 0.99 Yeo-johnson transformation gave best results

Python workflow

Exploratory analysis in Python

In both the training and validation set, there appears to be correlation variables Variation is greater in certain variables

Preliminary model comparison Python

Final results (Python and R)

Visualized final predictions as a binary heat map. Out of the 100 test samples, 99 samples were curves were classified correctly--with models in close agreeance.

Best performing models (in no particular order):

• Random Forest
• Naive Bayes
• XGBoost Classifier
• Gradient Boosting Classifier
• AdaBoost
• K-Nearest Neighbor
• Voting Classifier
• Bagging Classifier

Less well performing models(with more tuning better models may be possible):

• Decision Tree (would have been nice and interpretable)
• Support Vector Machine Classifier
• Log Regression