search

SrijanShovit / HealthLearning

A repo comprising of various Machine Learning and Deep Learning projects in healthcare domain.
36 stars 51 forks source link

Metabolic Syndrome Prediction | 5. Model prediction and evaluation #156

Open Arihant-Bhandari opened 3 weeks ago

Arihant-Bhandari commented 3 weeks ago

closes issue #98 Metabolic Syndrome Prediction | 5. Model prediction and evaluation

worked on step 5:

  1. Write generic functions.

  2. Use all features; no feature engineering (transformation, selection, extraction, dropping here)

  3. Use both normalized and standardized data.

  4. Use hyperparameter tuning for models.

  5. Use some training strategy like LOSO or K-Fold.

  6. Record time to find best params and time to train using those params.

  7. on test data and print all eval metrics.

  8. If it is binary classification, plot ROC to find optimal threshold.

  9. Using that, plot Confusion matrix to evaluate the models.

  10. Models to use: Extra Trees, SVM, Logistic, KNN, DT, RF, GB, Bagging Clf, XGB, Hist GB, MLP, Catboost, Adaboost, Naive Bayes, LightGBM

Worked on the following models:

1) XGBoost Classifier 2) Random Forest Classifier 3) Decision Tree Classifier 4) AdaBoost Classifier 5) LGBM (Light Gradient Boosting Machine) Classifier 6) Logistic Regression 7) Extra Trees Classifier 8) Histogram-based Gradient Boosting Classifier 9) Gradient Boosting Classifier 10) KNN (K-Nearest Neighbors) 11) Stacking Classifier 12) Voting Classifier 13) Bagging Classifier 14) CatBoost Classifier 15) Linear SVC (Support Vector Classifier) 16) SVC (Support Vector Classifier) 17) Perceptron 18) Multilayer Perceptron (Feedforward ANN) 19) Artificial Neural Networks

Arihant-Bhandari commented 3 weeks ago

hi @SrijanShovit this is the initial work i hope to get feedback on, from the problem statement i have devised stuff for XGBClassifier, if this work is found satisfactory i will continue with this for all other models, pls check up on this. if anything is to be added or changed, i will be glad to do them.

thank you for your time and patience.

github-actions[bot] commented 2 weeks ago

This issue has been automatically closed because it has been inactive for more than 7 days. If you believe this is still relevant, feel free to reopen it or create a new one. Thank you!

Arihant-Bhandari commented 3 days ago

Hi @SrijanShovit , the work on this PR is complete, pls review and if any suggestions or changes are to be made, please let me know, will gladly make the required changes.