Sunday, March 18, 2018

Predictive analysis for a telecom company

Is it possible to gain intelligence / impart intelligence  make a machine learn from the data and predict something ?
The current technology gives directions to this. Today with the computing capability of the latest machines/ highly advanced software  tools / development of human brain makes it possible.

Supposing we have a data set to understand the churn behaviour of  the customer from one service provider to another.
For instance we have  the following data about the customer
State
Eve Calls
Account Length
Eve Charge
Area Code
Night Mins
Phone
Night Calls
Int'l Plan
Night Charge
VMail Plan
Intl Mins
VMail Message
Intl Calls
Day Mins
Intl Charge
Day Calls
CustServ Calls
Day Charge
Churn?
Eve Mins


The first objective is to predict whether the customer will sustain with the telecom provider.The second objective is to understand which category of customers will tend to churn.

Where will we start?
1. First understand the type of data. Account length -say in days-- so integer - continuous variable
2. Classify what kind of variables these belong to-->  continuous, nominal, categorical, ordinal etc
3. Fix the target variable- in our case it is the churn- > Yes / NO.
4. Get the statistics of the churn customers.

5. Fix the drivers- ie the variables which lead to the decision.
Look at the example.

In the above , the four variables are narrowed down and fixed as drivers.

6. Use suitable alorithms to solve the problem. A typical algorithm that can be used is Classification and regression tress- C & RT / CART algorithm.
7. SPSS Modeler has an option to make the tree grow interactively.  Starting with the major driver- (from any of the above variables), and drilling down further  can give lot of insights to the problem.

No comments:

Post a Comment