The document describes using linear regression and PHP to optimize media buying by predicting the optimal cost and revenue based on campaign data from Google Analytics, cost, and revenue data from native ad networks. The algorithm uses linear regression to model the relationships between cost and clicks and revenue and clicks. The PHP application loads the data from a CSV file, calculates the correlation coefficients, and uses the linear regression models to predict the optimal cost and revenue for a given number of clicks.

1. Whirl Data
Media Buying Optimization
Using Linear Regression & PHP
Date: 30 June, 2016
By: Manivannan, Data Engineer

2. Whirl Data
Content
1. Objective
2. Algorithm
3. Implementation
4. Findings
2

3. Whirl Data
Objective
Case description:
I have a dataset that consists of campaign data from google analytics, cost and
revenue data. I would like to develop a model to optimize for efficient media
buying by optimizing the data from this dataset. The data is from native ad
networks (outbrain/mgid/taboola)
3

4. Whirl Data
Algorithm
For this case, we suggest Linear Regression for predicting the optimum
cost/revenue model.
4

5. Whirl Data
Implementation
We’ve created PHP web application to Extract the data from CSV file and apply
linear regression to solve the problem.
1. Load the data from CSV
2. Find the Correlation Coefficient of COST & CLICKS
3. Find the Correlation Coefficient of REVENUE & CLICKS
4. Predict the optimum cost for the given clicks
5. Predict the optimum revenue for the given clicks
5

6. Whirl Data
Find the Linear Regression of two fields(eg:Clicks and Cost)
Get the Coefficients of cost and clicks
Find the predict value of this Cost
For ex:
The optimum cost for 12 click is: 0.579792099792104
Optimum Cost for the Clicks
6

7. Whirl Data
Optimum Revenue for the Clicks
Find the Linear Regression of two fields(eg:Clicks and Revenue)
Get the Coefficients of Click and Revenue
Find the predict value of this Revenue
For ex:
The optimum Revenue for 12 click is:0.08521982711456341
7

8. Whirl Data
(Cost,Revenue and Click)
First find Linear Regression of Cost and Click
Second find the Linear Regression of Revenue and Click
Get the coefficient from first and second result
For ex:
The optimum Click for 0.3865 Cost is:7.9994
The optimum Click for 0.0284 Revenue is:3.999
8

9. Whirl Data
Findings
Linear regression models
9

10. Whirl Data
Application Screenshots
10

11. Whirl Data
Thank you
11