Distance Metrics

Best Practices

• What decisions do I need to make regarding my data? How might these decisions affect overall performance?

• Which predictors do I need? How can I confirm that I have the right predictors?

• What parameter values (if any) should I choose for my model? How can I find the optimal value for a given parameter?

• What metrics will I use to evaluate the performance of my model? Why?

• How do I know if there's room left for improvement with my model? Are the potential performance gains worth the time needed to reach them?

Workflow

• First

  • import standard libraries

  • import and read dataset

• Preprocessing Data

  • Remove unnecessary columns

  • Convert feature(s) to binary encoding

  • Detect and deal with null values

  • One-Hot Encode categorical columns

  • Store target column in a separate valiable and remove it from DataFrame

• Normalize Data

  • StandardScaler

  • .fit_transform()

  • Creating Training and Testing Sets (train_test_split)

• Creating and Fitting KNN Model

  • KNeighborsClassifier

  • Fit the classifier to training data/labels (labels = target)

  • Use the classifier to generate predictions

• Precision, Recall, Accuracy and F1-Score

  • from sklearn.metrics import precision_score, recall_score, accuracy_score, f1_score

• Improving Model Performance

  • take in six parameters:

    • X_train, y_train, X_test, and y_test

    • min_k and max_k. Set these to 1 and 25, by default

    Create two variables, best_k and best_score

    Iterate through every odd number between min_k and max_k + 1.

    For each iteration:

    • Create a new KNN classifier, and set the n_neighbors parameter to the current value for k, as determined by our loop.

    • Fit this classifier to the training data.

    • Generate predictions for X_test using the fitted classifier.

    • Calculate the F1-score for these predictions.

    • Compare this F1-score to best_score. If better, update best_score and best_k.

    Once it has checked every value for k, print out the best value for k and the F1-score it achieved.