Skip to main content

Predictive Model (Using Trend Analysis, Univariate Analysis and Binomial Logistic Regression)


Predictive models have been in use since advent of various statistical tools. Both marketing and financial institution uses it extensively. This article tries to look under the hood how a predictive model is created. I will try explaining it using simplest available statistical tool like Trend analysis, Univariate Analysis and slightly complex tool called Binomial Logistic Regression.
Lets assume company want to know how much attrition will happen in current quarter, so that it can start hiring process.
First step of model is find out what are the factors that can lead to attrition of an employee. To find these factors we simple take past data and holistic list of factors.


Age, Income, Martial Status, Education Background, Work Rating, Number of leaves, Number of days came later than usual timing, Number of year in company, Business Unit, Business vertical, Past Work experience

Indicative list of factors for Company Attrition Model

On each listed factor we do trend analysis and Univariate analysis. Tread analysis will give us a rough relation between each factor and attrition. In our case we will find out trend between attrition and age. Outcome can be something like shown below.

Sample Trend Analysis Result
(Attrition of Employee less than 26 Year in age against number of quarters spent in Company)

Similarly once done for all parameters, researcher will get an idea, which are the parameters that are relevant for prediction. In this process of selecting the parameters one should always ask, if the result shown by graph makes sense, if it defies normal course of happening or observation, a researcher should try to look into the data, verify its validity.

Once all the parameters are chosen with help of Univariate and Trend analysis, its time to create the required model. For this purpose we divide available past data into two parts. 2/3 of randomly selected data is used to create the model and rest 1/3 will be used to verify the model.

Prepare 2/3 randomly selected data for binomial logistic regression. One can use any analytics tool like SPSS.

Once we get the parameter coefficient from regression analysis, we use this equation to predict the event. We test the model using 1/3 data kept aside for testing the model. A good fit will give you your model, otherwise one need to go back to parameter discovery exercise once again and repeat the whole process.  

Predictive model has innumerable usage and if created with care, it can help in streamlining many complex process.



Labels:

Comments

Post a Comment

Popular posts from this blog

Open Source Architecture for Building BPM Solution - Part 1

[This article talks about how a custom BPM solution can be build using available open source BPM tools like Activiti or jBPM] Business Case: A MFI is looking to automate its processes. It has basic functions called loan origination and loan servicing Loan origination includes following steps a) Borrower risk analysis b) Borrower on-boarding where as loan servicing includes a) Disbursements b) Collection c) Closer d) Loan extension etc. Other than these two functions MFI also have operations team which takes care ground level activities like physical transportation of cash and documents, deployment of agents, identification of loan potential areas. So how can we build solution which is scalable and minimal cost for this MFI? Below is one of many solutions Step 1: High Level Architecture Design Step 2: Process Discovery and Implementing the Executable processes Lets restrict ourselves to only loan servicing – disbursements of funds. And see how we can fit that in ...
Post a Comment
Read more

How Kirana stores can handle Organize Retail in India

Organize retail formats include departmental stores, hypermarkets, supermarkets and specialty stores. Any store having more than 500 sq. feet is counted in organized retail. There are more than 14 million retail shops in India, organized retail accounts for only 4-5% [i] , but it is growing at 40% per annum. Organized retail in India will grow faster as regulations loosen up. Government will sooner or later give in to demand of better FDI regulations, in favor of big corporate houses. Time is not far away when Kirana stores will walk the path of obsolescence. So how can these small stores save themselves? SWOT Analysis Kirana Stores Organized Retail Stores Strength Well connected to community, Customized service, Near to customer Bargaining Power, Allows customer to choose, Volume Sales, Financial strength, Bigger display area, Trendy Weakness Outdated, relatively costlier, Less display area, Higher cost per SKU ...
Post a Comment
Read more