Then finance and banking industry has gone through massive shifts in the in the last decade. Between the downturn from the financial crisis in 2008 and the increasingly disruptive pressure from Silicon Valley it is even more important to understand our customers’ behavior. (Deloitte, 2016) Direct sales marketing is one of the most effective methods for both gaining new customers as well as increasing the deposits of our existing customers. Globalytics Capital needs to gain better insight into its customer’ behavior to be poised for success in the future. In order to position Globalytics Capital the Data Access Response Team (D.A.R.T) is proposing an analysis be done using the customer marketing database.

Analysis Method

As a proof of concept D.A.R.T. has created this report detailing how such an analysis could be conducted using a similar banking database. The data that D.A.R.T will be utilizing comes from the Center for Machine Learning and contains the result of direct marketing campaigns from another banking institution. (Moro, Cotez, & Rita, 2014) Globalytics Captial conducts similar campaigns throughout the year. A direct marketing campaign consists of phone calls made to potential customers with the positive outcome being a deposit made with the bank. The data set that we will be looking at contains 17 input attributes and one output attribute. The input attributes are organized in 3 groups.

1. Bank Client Data
2. Last Contact Data
3. Extended Contact Data

Bank Client Data

This is demographic data about the customer being contacted. The information captured here helps the marketer better understand the customer’s life circumstances and whether they have existing loans. The following table details these attributes.

Last Contact Data

This is data about the last contact with the customer during the current campaign. It is important to know how long it has been between customer contacts. Too frequent contacts can result in negative outcomes with customers so marketers need to know this information.

Extended Contact Data
This data gives us additional information about the current campaign as well as the previous marketing campaign. This is important to understand how this customer has responded in the past.

Finally, there is the output variable labeled y which shows whether the call was successful.

Data Mining Approach

For this example, the apriori algorithm will be used to determine the which customers will be most likely to make a deposit during a marketing campaign. This will allow us to prioritize the customers during the campaign. Several of the fields within this data set will need to be modified before any processing can being. Six of the fields will need to be changed from continuous numeric fields into discrete bins. Specifically, the age, balance, duration, campaign, pdays, and previous fields need to be modified. This allows us to count the number of rows. The below chart shows the age range pins.

Interpretation

Within the data set there were a total of 4521 calls made during this campaign. 521 calls were successful while 4000 were not successful resulting in a 11% success rate for this campaign. This again shows the importance for knowing which customers to target.

After interpreting the data, it was determined that there were only 2 conditions that have an impact on the outcome of a solicitation call. Customers with no personal loans and nothing in default are the most likely to respond favorably. For both of these conditions the algorithm returned with around 12% confidence. This is not a high level of confidence however this is in line with the 11% success rate for the campaign.

Conclusions

Moving forward the team feels that we should adjust the bins for some of the continuous fields. This would allow for rule sets with a higher support and confidence levels. The data set was a small sample of the data that would be available from Globalytics Capital. With a larger data set there would be a greater opportunity for building association rules.

References

• Deloitte. (2016, July 1). Banking Industry Outlook: Banking reimagined. Retrieved Oct 17, 2016, from Deloitte: http://www2.deloitte.com/us/en/pages/financial-services/articles/banking-industry-outlook.html
• Moro, S., Cotez, P., & Rita, P. (2014, June 1). Bank Marketing Data Set. Retrieved from Machine Learning Repository: https://archive.ics.uci.edu/ml/datasets/Bank+Marketing

Apendix A – R Code

library(arules) library(arulesViz)
setwd("~/Documents/DATA 630 FALL 2016/Assignment 3/bank")

bank <- read.csv("~/Documents/DATA 630 FALL 2016/Exercise 3/Bank.csv") str(bank)

bank$age <- discretize(bank$age, "frequency", categories=6) bank$balance <- discretize(bank$balance, "frequency", categories=6)
bank$duration <- discretize(bank$duration, "frequency", categories=6)
bank$campaign <- discretize(bank$campaign, "frequency", categories=6)
bank$pdays <- discretize(bank$pdays, "frequency", categories=6)
bank$previous <- discretize(bank$previous, "frequency", categories=6)

rules<-apriori(bank, parameter=list(supp=0.1, conf=0.1, minlen=2), appearance=list(rhs=c("y=yes"), default="lhs"))

rules.sorted <- sort(rules, by="lift")

subset.matrix <- is.subset(rules.sorted, rules.sorted) subset.matrix[lower.tri(subset.matrix, diag=T)] <- NA
redundant <- colSums(subset.matrix, na.rm=T) >= 1
rules.pruned <- rules.sorted[!redundant]

plot(bank$age, col=rainbow(6), xlab="Age Range", ylab="Number of Customers", main="Count per Age Range")

lbls <- paste(names(table(bank$y)), "\n", table(bank$y), sep="") pie(table(bank$y), col=rainbow(4), main="Deposit Made", labels=lbls)

Summarize what data mining tools can reveal about the Iris data.

Data mining tools can help to identify patterns and trends within the Iris data set. These types of tools can also facilitate statistical analysis of the data. Tools like R can be used to explore that data and gather descriptive statistics on the data such as record count, mean, and percentiles. Through this identification of patterns machine learning models can be built and trained. The Iris dataset was used during the recent Google Developers series on Machine Learning. (Gordon, 2016)

Why is it important for the data-mining tool to have visualization capabilities?

Data visualization is the process of showing information in a tabular or graphical way. (Tan, Steinbach, & Kumar, 2006) While spreadsheets are one way of visualizing data it is often more useful for an audience to see data represented in a graphical way. Charts and graphs can be used to visualize patterns in the Iris dataset. One of the reasons that a tool like R is powerful are the built-in visualizations. Instead of having to transpose data from one tool into another you can build pie charts, histograms, and scatter plot graphs directly in the tool. This eliminates the data entry errors that can occur across tools.

Is an open source data mining tool better than a commercial one?  Why or why not?

Over the course of my career I have worked with both open source and commercial software tools. Open source software described software whose source code is available for free redistribution and can be modified by anyone. (Open Source Initiative, 2007) Using open source software and tools typical has a lower cost of ownership that commercial tools. These types of tools often include a community of users who volunteer best practices and feature upgrades. This can give an organization greater flexibility over the control of their tools. The downside to using an open source tool comes in resolving bugs when an open source tool is abandoned. Since these types are tools are supported by volunteers many projects are abandoned with no avenue for getting fixes. On the other hand, commercial tools typically have dedicated support infrastructures. Enterprise customers purchases legally binding service level agreements(SLAs) that promise both timely support and upgrades. With the support systems behind them commercial tools have much higher cost of ownership for an organization. As a systems engineer for the federal government we use a combination of both commercial and open source software and tools. I encourage my clients to decide on which to use on a case by case basis. For the benefit of this question I would recommend using open source tools initially. The low cost of ownership and flexibility allows you to learn. Once the project has matured or needs a greater level of support than it is time to examine moving to a commercial tool.

References

Gordon, J. (2016, March 30). {ML} Machine Learning Recipies. Retrieved May 19, 2016, from Youtube: https://www.youtube.com/playlist?list=PLKX8hZ1Cat4P2cG_ha40e803oOl4aM7MK
Open Source Initiative. (2007, March 22). The Open Source Definition. Retrieved May 19, 2016, from Open Source Initiative: https://opensource.org/osd
Tan, P.-N., Steinbach, M., & Kumar, V. (2006). Introduction to Data Mining (Vol. 3). Boston, MA, USA: Pearson Education, Inc.