From late April to early May, I got to participate in the Cortex Analytics Simulation Competition by SAS in collaboration with HECMontréal. The competition was open to everyone, especially targeted to students, who wanted to explore how predictive analytics is used in a real-world scenario.
I won’t go into too much detail about the game since I don’t want to give away possible answers for future competitions, but I would say it was a good experience learning about predictive analytics using SAS Enterprise Miner.
If you come across the competition and would like to learn more, this article goes through the game scenario, instructions and my personal experience throughout the competition.
You’re working on a fundraising campaign for a not-for-profit charitable organisation with 1 million members. The foundation decided to add a direct contact campaign to its list of marketing activities. The objective is to fundraise the most donations given the costs of calling members. This will be done using predictive modeling software (SAS Enterprise Miner) to predict how many and which individuals to target in the campaign.
You will be provided with the dataset of potential donors. You will fit different models based on the previous behavior of donors, scoring donors to predict the donation this year. Using this output, decide how many potential donors to target. Upload your list of IDs which will be ranked from the submissions based on operating surplus: sum of donations - total cost of calling.
I created a table of some of the information given below.
Table 1: some variables listed, Table 2: cost of contacting members
As it’s an introductory competition, participants are given detailed game instructions and could attend daily 1-hr webinars (provided by SAS).
These webinars included presentations going through the instructions and Q&A sessions. We were given two different methods of modeling:
Of course, the 2-stage modeling process was more effective. It is more commonly known as uplift modeling, where the uplift is calculating by computing the difference between the predicted amounts which take into account the probability of giving when a person is contacted or not.
#modeling #data-science #machine-learning #review #predictive-analytics #deep learning