Suppose you are training a classifier with 2000 data points belonging to a positive class and 250 belonging to a negative class.

1. Which performance metric would be adequate to assess the performance of the model?
2. Which performance metric would be inadequate for this data set?

Question

Suppose you are training a classifier with 2000 data points belonging to a positive class and 250 belonging to a negative class.

1. Which performance metric would be adequate to assess the performance of the model?
2. Which performance metric would be inadequate for this data set?

BlackTom AI · Accepted Answer

When evaluating a binary classifier on an imbalanced dataset, the choice of metric matters a lot because accuracy can be misleading if one class dominates.
Option 1: 'Precision; MCC' – Precision measures the positive class performance, but using just precision (without considering negatives) can give an inflated sense of performance if the model skew favors the positive class; MCC is a balanced measure, but the pairing here is inconsistent with the question asking for which metrics are adequate/inadequate as a pair.
Option 2: 'MCC; Accuracy' – This pair contrasts a balanced metric (Matthews Correlation Coefficient) with overall accuracy. MCC is generally considered suitable for imbalanced datasets because it accounts for true/false positives and negatives in a balanced way, whereas accuracy can be misleading due to class imbalance (2000 positives vs 250 negatives). This pairing captures the idea that MCC is adequate while accuracy is inadequate in such skewed settings.
Option 3: 'Accuracy; Recall' – Here, accuracy is the more problematic metric in an imbalanced context, and recall (which focuses on the positive class) does not address performance on the negative class at all. This makes the pair inappropriate for describing adequate/inadequate metrics in this scenario.
Option 4: 'Recall; MCC' – While MCC is indeed a good choice for imbalanced data, recall alone is not an adequate measure of overall classifier performance because it ignores how well the model identifies negative instances; the combination suggests recall is adequate, which is misleading in this context.
In summary, the most sensible pairing respects the known strengths and weaknesses of the metrics under class imbalance: MCC is a robust, balanced metric, whereas accuracy can be misleading due to the difference in class sizes. The other options either rely on a metric that ignores important error types or pair metrics in a way that mischaracterizes their informativeness for imbalanced data.

COMP90089_2025_SM2 Quiz 3: Supervised Learning

Suppose you are training a classifier with 2000 data points belonging to a positive class and 250 belonging to a negative class. 1. Which performance metric would be adequate to assess the performance of the model? 2. Which performance metric would be inadequate for this data set?

查看解析

登录即可查看完整答案

类似问题

Which of the following are appropriate ways to evaluate the performance of a Support Vector Machine (SVM) used for classification? (Select all that apply)

Which indicators are used for model evaluation in classification? (Select all that apply)

Which indicators are used for model evaluation in classification? (Select all that apply)

Match the contents to their correct groups. 1: True positive rate (TPR) 2: True negative rate (TNR) 3: Precision 4: Area under the curve (AUC)

Question at position 7 True/False Question: The accuracy metric can distinguish between the various types of misclassification errors a classifier can make. TrueFalse

更多留学生实用工具

考试浏览器助手

风格化写作助手

论文查重助手

文献引用助手

课堂转译助手

课堂笔记助手

Quiz搜索助手

学校历年真题

智能刷题助手

智能匹配练习

希望你的学习变得更简单