Quiz collegato alla guida · Medio Livello

Cost-Sensitive Learning Quiz

Covers cost matrices, weighting and resampling choices, and the assumptions behind a decision threshold that minimizes expected error cost.

Percorsi guida correlatiCost Sensitive Learning
Domanda 1 di 8

Quale presupposto fondamentale rimuove l’apprendimento sensibile ai costi dalla formazione standard basata sull’accuratezza?

Anteprima di ogni domanda8 domande

Le risposte e le spiegazioni vengono visualizzate mentre rispondi al quiz.

  1. Quale presupposto fondamentale rimuove l’apprendimento sensibile ai costi dalla formazione standard basata sull’accuratezza?
    • That models must use gradient descent
    • That data must be labeled
    • That every type of misclassification carries the same cost
    • That features must be numeric
  2. Nell’esempio del rilevamento delle frodi, perché una frode mancata viene penalizzata più pesantemente di un falso allarme?
    • Because false alarms are illegal
    • Because fraud is always rarer than legitimate transactions
    • Because the model cannot detect false alarms
    • Because an undetected fraud costs far more than a customer service call for a false alarm
  3. Secondo la guida, l’apprendimento sensibile ai costi equivale alla gestione dello squilibrio di classe?
    • No, they overlap but are distinct, since a balanced dataset can still need cost-sensitive treatment
    • Yes, they are always identical
    • No, cost-sensitive learning only applies to text data
    • Yes, because imbalance always implies unequal costs
  4. Quale parametro menziona la guida che librerie come scikit-learn o XGBoost utilizzano per implementare direttamente la ponderazione delle classi?
    • learning_rate
    • class_weight or scale_pos_weight
    • max_depth
    • n_estimators
  5. Quale tecnica utilizza MetaCost, come descritto nella guida?
    • It removes all minority class examples
    • It only adjusts the learning rate schedule
    • It ignores the cost matrix entirely
    • It relabels training examples based on expected cost before fitting a standard classifier
  6. Under which assumptions does the guide’s C(FP)/(C(FP)+C(FN)) threshold apply?
    • Any score output, even if uncalibrated
    • Only when both classes are equally common
    • Calibrated posterior probabilities, two classes, zero correct-decision costs and fixed error costs
    • Only after resampling the training set
  7. Why does the guide describe threshold adjustment as often the simplest cost-sensitive technique?
    • Because it requires retraining the entire model
    • Because it requires no retraining, only recalibrating the decision boundary using existing predicted probabilities
    • Because it eliminates the need for a cost matrix
    • Because it only works on image data
  8. What does the guide identify as the main practical bottleneck limiting wider adoption of cost-sensitive learning?
    • Stakeholders being willing and able to specify costs accurately
    • Lack of available programming libraries
    • Insufficient computing power
    • Legal bans on cost matrices