You are on page 1of 2

Figure 3.

5 Linear regression[11]

Suppose we have data X, and the target values Y. “The aim of linear regression algorithm is to
train a linear model which can be used to predict a new Y given” an unseen X with the
minimum amount of error as possible. We need to think about how much salary somebody earns
depending on how long of instruction they got. For this reason, we use gradient descent[11] .

“The goal of gradient descent method or algorithm is to take the model’s loss function and find
its minima and use it to get the model parameters in a better state as compared to the previous
state iteratively. Each iteration being better than the previous one in terms of loss function. A
common problem in all of machine learning is overfitting[11]. It can be defined as learning the
outputs to particular inputs so that they can provide correct answers when the same example
comes but is not so good in case a never seen before data item comes. It happens when a model
learns so much from the training data that it finds relations which aren’t actually present in the
data but by coincidence fit in the training data. As the” "model grows more and more complex,
this problem goes more and more prevalent. Opposite of it is underfitting which is an issue when
the model doesn’t learn much of the relations between input and output and performs poorly on
both training and testing data.”

3.3.3.1.2. Polynomial Regression

Polynomial Regression[13] is similar to Linear Regression, except that it uses a curve that is a
polynomial function of the inputs and their powers, instead of just a linear function of the inputs.
For example, an output Y is to be predicted using input parameters X. “If the output Y is a linear
function of input X, linear regression would work well. However, if it is not a linear function of
input X, the model will underfit[13] the data as shown in figure 3.6.”

Figure 3.6 Linear Model Underfitting Data[13]

However, “if a polynomial model is fitted to the given data, then it fits the data more accurately
with underfitting and hence it improves the prediction of the model. Thi

You might also like