Performance metric Problem $(15$p$)$

Let's imagine we develop a model with the goal of determining if an email is Spam or Not

Spam. Once the model has been trained, we use it on a test group of $500$ fresh emails that

have also been labeled. The model then generates the subsequent contingency table.

a$)$ What is the accuracy $?(5$p$)$

b$)$ Do you think that accuracy is a good measure? Why or why not? $(5$p$)$

c$)$ If we increase size of training set, would this improve or deteriorate the

performance of the model? $(5$p$)$