a$)$ What are the possible reasons for model overfitting?

b$)$ What can you do to avoid overfitting? Elaborate you answers with decision tree models and

logistic regression.

c$)$ Why accuracy can be a bad metric to evaluate your classifier? What other metrics you can

use?

d$)$ Consider a synthetic data set which is generated from the function $sin(2x),$ which is shown by

the green curve in Figure $2($a$).$ Let us assume that we decided to fit a polynomial function of

the following form:

$y(x,w)=w_0+w_{1}x+w_2{x}^2+$dots$+w_M{x}^M$

Figure $2($b$)$ shows what happens when we fit a linear model $(M=1$ in the Equation$)$ to the

observations. Is this model a good fit? If not, is it underfitting or overfitting?

Figure $2($c$)$ shows what happens when we fit a higher order polynomial $($degree $=M=9$

in the Equation$)$ to the observations. Is this model a good fit? If not, is it underfitting or

overfitting?