Understanding Important Features in Marketing Campaign Analysis
This morning I encountered an article about marketing campaign data analysis using “correlation funnel”, which is basically a plot(visualization) that lists important(highly-correlated) variables to understand/predict a target variable.
To use correlation funnel, you have to go through some data wrangling process and perform binary correlation analysis using R; anyhow, the plot shows all the correlated variables from the highest correlation to the lowest in a single chart, the major benefits could be:
- Speeds Up Exploratory Data Analysis
- Improves Feature Selection
- Gets You To Business Insights Faster
And I downloaded the same dataset and compared the correlation funnel chart with HEARTCOUNT’s difference analysis chart:
Dataset Information
- The data is related with direct marketing campaigns of a Portuguese banking institution. The marketing campaigns were based on phone calls.
- go to the site and click "Data Folder" to download the dataset: https://archive.ics.uci.edu/ml/datasets/Bank+Marketing
- for this article, I am using a "bank-full.csv" dataset.Call Duration (numeric/continuous variable)
Marketing campaign analysis often involve a response variable (whether customers responded with the campaign or not) and other predictors (age, sex, marital status, psychological/behavioural traits, etc.). Our goal is to determine which predictors are important in understanding or predicting customer’s response behaviour(yes or no).
First, call duration is the highest correlated feature with the target variable [response=yes or no] in both tools.
- “correlation funnel” says [call duration >319] is highly correlated with [response=yes].
- HEARTCOUNT says [301 < call duration < 2810] is the critical call duration if we wish to get higher [response=yes; shown as Group B below]. (in fact, when call duration is longer than 2810 secs, response ratio actually gets worse)
Housing and Month(categorical variables)
In case of [housing] and [month] variables, the results of two tools are similar. However, in HEARTCOUNT, the user can have quantitative information backing why those with [housing=yes] are less likely to respond with the campaign. (58% of [response=no] had [housing loan=yes] while only 36% of [response=yes] had [housing loan=yes])
The tool we use shapes our understanding of the world. If you wants to download HEARTCOUNT use cases in marketing campaign analysis and customer segmentation, please click the link
