Cuestionario vinculado a una guía · Medio Nivel

Prueba del coeficiente de correlación de Matthews

Pruebe el razonamiento de matrices de confusión, los cálculos de MCC y los obstáculos del desequilibrio de clases con ejemplos de filtros de spam y detección de defectos.

Rutas de guía relacionadasCoeficiente de correlación de Matthews
Pregunta 1 de 8

A filter labels all 1,000 messages legitimate when 990 truly are legitimate. Why is its 99% accuracy insufficient evidence of useful spam detection?

Vista previa de cada pregunta8 preguntas

Las respuestas y explicaciones aparecen a medida que realiza el cuestionario.

  1. A filter labels all 1,000 messages legitimate when 990 truly are legitimate. Why is its 99% accuracy insufficient evidence of useful spam detection?
    • Every majority-class error receives more weight than a spam error
    • Accuracy alone establishes that both classes have high recall
    • The filter has detected none of the ten spam messages
    • An MCC of 0.99 follows directly from this accuracy
  2. Which value results from the guide's toy confusion matrix with 40 true positives, 40 true negatives and 10 of each error?
    • MCC of 0.6
    • MCC of 0.8
    • MCC of 0.4
    • MCC of 1.0
  3. A report interprets an MCC of 0.6 as '60% of cases were correct.' How should that interpretation be corrected?
    • It means 60% of positive cases were found
    • It means 60% of predicted positives were correct
    • It means the model is calibrated at 60% confidence
    • MCC is a correlation measure, not the percentage classified correctly
  4. For binary data containing both classes, a classifier reverses every label perfectly. Which MCC value describes that relationship?
    • 0
    • -1
    • 0,5
    • 1
  5. A factory values missed defects differently from unnecessary inspections. Which additional evidence is needed alongside MCC?
    • A single pooled count of all mistakes, without their types
    • Only the accuracy at the current operating threshold
    • The error counts and their operational costs
    • Only a probability-calibration plot, without error counts
  6. Which inputs are appropriate for scikit-learn's matthews_corrcoef in the workflow described?
    • Observed labels and predicted labels
    • Observed labels and unthresholded probability estimates
    • Predicted probabilities and their rounded copies
    • Training labels and labels from unrelated evaluation cases
  7. Why does predicting a single class for every case require care when interpreting the binary MCC formula?
    • The score must equal the majority-class accuracy
    • The absent predicted class can be ignored without affecting the formula
    • Perfect recall for the chosen class makes the denominator one
    • One denominator factor becomes zero
  8. When extending an MCC report to three classes, which approach matches the guide?
    • Compute three binary MCC values and take their unweighted mean
    • Use a multiclass MCC implementation and inspect the confusion matrix
    • Merge the two rarest classes before interpreting the original task
    • Use overall accuracy as the multiclass definition of MCC