The figure below illustrates two different models: Model A (dotted line) and Mod

Question

The figure below illustrates two different models: Model A (dotted line) and Model B (dashed line). The models vary in complexity. Both models have been applied to the same training dataset (circles) that appears to be characterized by a quadratic function (solid line). Recall that the expected error of a classification model can be decomposed into three components: bias, variance, and noise. Use the figure below explain the three errors. In particular, identify which of the models is more likely to suffer from high variance and which is more likely to suffer from high bias. Also, discuss the most likely performance of the Models when they are applied to new records.

Explanation / Answer

a) Explaining through Model A:

Bias: The difference between true function(dark line) and average of h(x*) is the bias i.e [f(x*) - hm(x*)]

which describes the average error of h(x*)

Variance: It tell us how much h(x*) vary from one training set to another

it is E[h(x*) - hm(x*)^2]

Noise: It tells us how much y* varies from f(x*) which is nothing but the errors

From the definitions given above it is very easy to tell that, For Model A, Bias is more, but for Model B the variance will be more

So when they are applied to new records, there is a chance that B cannot hold good in those new records, A can. SInce B is expected to have more variance, this is going to happen, However, if the new records has the same trend as the training dataset, then as a whole Model B can be stated as a better model than model A

Related Questions

Navigate

• Browse (All)
• Browse T
• Subjects

---

---

Integrity-first tutoring: explanations and feedback only — we do not complete graded work. Learn more.