Question6 Suppose that you have used a model to do a binary classification task where 50% of the data is from class 1 and the rest from class 2. Your training accuracy is around 90% and your validation accuracy is around 60%, how you interpret the result and what would be your next action? (select one) The model is overfitting the data and you will reduce the complexity of the model or increase your training sample The model is overfitting the data and you will increase the complexity of the model or increase your training sample The model is underfitting the data and you will reduce the complexity of the model or increase your training sample The model is underfitting the data and you will increase the complexity of the model or increase your training sample ResetMaximum marks: 1.5 Flag question undefined

Question

BlackTom AI · Accepted Answer

Examining the scenario: a binary classification with class balance (50/50) where training accuracy is ~90% but validation accuracy is only ~60%. This pattern strongly suggests that the model has learned the training data too well, including noise and idiosyncrasies, but fails to generalize to unseen data. This is characteristic of overfitting, not underfitting.

Option 1: 'The model is overfitting the data and you will reduce the complexity of the model or increase your training sample' This aligns with the diagnostic signal (high training accuracy, much lower validation accuracy). Reducing model complexity (e.g., fewer layers/parameters, simpler architecture) or adding more training data can help improve generalization.

Option 2: 'The model is overfitting the data and you will increase the complexity of the model or increase your training sample' Increasing model complexity would typically worsen overfitting and further hurt validation performance, unless there is a mistaken belief that more capacity will generalize better; generally this is not the recommended corrective action for overfitting.

Option 3: 'The model is underfitting the data and you will reduce the complexity of the model or increase your training sample' Underfitting would show poor training and validation accuracy; here training accuracy is high, so underfitting is unlikely. Reducing complexity would worsen underfitting further.

Option 4: 'The model is underfitting the data and you will increase the complexity of the model or increase your training sample' While increasing complexity can help reduce underfitting, the given evidence points to overfitting, not underfitting, since the model already fits the training data well.

Additional nuance: With a balanced dataset and a validation drop from 90% to 60%, the gap indicates poor generalization rather than simply needing more data in all cases; sometimes techniques like regularization, dropout, or cross-validation can help, alongside reducing capacity or augmenting data. Ensuring the evaluation setup is correct (no data leakage, proper shuffling) is also prudent.

In summary, the most consistent interpretation is that the model is overfitting, and the recommended next actions would be to reduce complexity or increase the training data. The other options either contradict the observed evidence (overfitting vs underfitting) or propose actions counterproductive to overfitting (increasing complexity).

查看解析

登录即可查看完整答案

类似问题

In the above image, the prediction made at any value of X is shown by the blue line. This predictive model is an overfit for the training data.

Which one of the following linear discriminants is most prone to overfitting a training data set?

Overfitting (1 mark)What is overfitting?[Fill in the blank]

Which of the following conditions can lead to logistic regression overfitting the training data?

Question at position 6 True/False Question: There is no direct relationship between model complexity and overfitting. TrueFalse

A classifier shows 100% accuracy on the training set but performs poorly on new cases. What is most likely true?

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