How It Works by Math

Mathematical explanation for Linear Regression working

Suppose we are given a dataset:

Given is a Work vs Experience dataset of a company and the task is to predict the salary of a employee based on his / her work experience. This article aims to explain how in reality Linear regression mathematically works when we use a pre-defined function to perform prediction task. Let us explore how the stuff works when Linear Regression algorithm gets trained.

Iteration 1 – In the start, $θ_0$ and $θ_1$ values are randomly chosen. Let us suppose, $θ_0 = 0$ and $θ_1 = 0$.

• Predicted values after iteration 1 with Linear regression hypothesis. $h_\theta = \begin{bmatrix} \theta_0 & \theta_1 \end{bmatrix} \begin{bmatrix} x_0 & x_0 & x_0 & x_0 \\ x_1 & x_2 & x_3 & x_4 \end{bmatrix}$ $= \begin{bmatrix} 0 & 0 \end{bmatrix} \cdot \begin{bmatrix} 1 & 1 & 1 & 1 \\ 2 & 6 & 5 & 7 \end{bmatrix} = \begin{bmatrix} 0 & 0 & 0 & 0 \end{bmatrix}$\

• Cost Function – Error $J(\theta) = \frac{1}{2m} \sum_{i=1}^{m} \left[ h_\theta(x_i) - y_i \right]^2$

$= \frac{1}{2 \times 4} \left[ (0 - 3)^2 + (0 - 10)^2 + (0 - 4)^2 + (0 - 3)^2 \right]$

$= \frac{1}{8} \left[ 9 + 100 + 16 + 9 \right]$

$= 16.75$

• Gradient Descent – Updating $θ_0$ value Here, j = 0 $\theta_j := \theta_j - \frac{\alpha}{m} \sum_{i=1}^{m} \left[ \left(h_\theta(x_i) - y_i\right) x_i \right]$

  $= 0 - \frac{0.001}{4} \left[ \left(0 - 3\right) + \left(0 - 10\right) + \left(0 - 4\right) + \left(0 - 3\right) \right]$ $= \frac{0.001}{4} \left[ -3 + (-10) + (-4) + (-3) \right]$ $= \frac{0.001}{4} \left[ 20 \right]$ $= 0.005$

• Gradient Descent – Updating $θ_1$ value Here, j = 1 $= 0 - \frac{0.001}{4} \left[ (0 - 3) 2 + (0 - 10) 6 + (0 - 4) 5 + (0 - 3) 7 \right]$ $= \frac{0.001}{4} \left[ -6 + (-60) + (-20) + (-21) \right]$ $= \frac{0.001}{4} \left[ -107 \right]$ $= 0.02657$ Iteration 2 – = 0.005 and $θ_1$ = 0.02657\

• Predicted values after iteration 1 with Linear regression hypothesis. $h_\theta = \begin{bmatrix} \theta_0 & \theta_1 \end{bmatrix} \begin{bmatrix} x_0 & x_0 & x_0 & x_0 \\ x_1 & x_2 & x_3 & x_4 \end{bmatrix}$ $= \begin{bmatrix} 0.005 & 0.026 \end{bmatrix} \cdot \begin{bmatrix} 1 & 1 & 1 & 1 \\ 2 & 6 & 5 & 7 \end{bmatrix} = \begin{bmatrix} 0.057 & 0.161 & 0.135 & 0.187 \end{bmatrix}$\

Now, similar to iteration no. 1 performed above we will again calculate Cost function and update $θ_j$ values using Gradient Descent. We will keep on iterating until Cost function doesn’t reduce further. At that point, model achieves best $θ$ values. Using these $θ$ values in the model hypothesis will give the best prediction results.\