1

BSAN2205 MACHINE LEARNING FOR BUSINESS

Project Plan

The course BSAN2205 Machine Learning for Business has three assessment items including a

Project Plan, a Project Report and Presentation, and a School-based Take-home Assessment

(weighted 20%, 50%, and 30%, respectively).

These notes outline my expectations for the

Project Plan and introduce the context for the project work.

I intend the Plan or proposal to be a

formative piece of assessment.

The Plan should set the groundwork for your project and project

report.

I will provide feedback on your Plan that you can incorporate into your project.

Background and Context

In competitive markets, businesses face the challenge of acquiring and retaining customers.

Consider subscription services, for example, subscriptions to digital editions of newspapers and

magazines, subscriptions to streaming services (film and television, music, news, sport, etc.), and

subscriptions to cable television services (Foxtel).

Other businesses face the same challenges, for

example, airlines, banks, insurance companies, telecommunication companies, and retailers,

restaurants, and personal services businesses.

One retention strategy is to deepen relationships

with customers through “upselling” – convincing a customer to buy something in addition to or more

expensive than that they have previously purchased from a business.

Streaming services like Netflix

and Spotify strive to build customer “engagement” – increasing the number of downloads and/or

the time spent streaming.

Bank marketing provides the specific context for the project.

Like many consumer businesses, banks

confront the challenges of attracting new customers and retaining existing customers.

Strategies for

retaining customers provides the setting for the project.

For banks, engagement is reflected in the

number of products (active accounts) customers maintain.

Often retention strategies have the goal

of deepening engagement by encouraging customers to open new accounts.

Consolidating accounts

with one rather than many banks may offer consumers some benefits at the margin.

For example,

highly engaged customers may be offered lower rates on loans, access to services for which they do

not have to pay (at least, not directly), and minimising the overall burden of managing multiple

banking relationships.

For banks, the benefits of more highly engaged customers are larger and

more stable cash flows, lower marketing expenses (with the costs of attracting a customer higher

than the costs of retaining a customer, per customer relationship economics), and thus potentially

higher profits.

Before moving on, I would like you to appreciate that in problems in business can be solved through

effective predictive models of binary outcomes.

The decision to purchase or not purchase shares in

a company, to acquire or merge with another business, to hire or not hire a prospective employee,

etc.

All of these decisions involve binary outcomes (in some cases, they can be characterised as

“go/no go” decisions).

The specific focus of the project is customer acceptance of a marketing offer,

but the concepts and models have much broader application.

Aims of the Proposal

The Project Plan has two broad aims.

Firstly, the Plan is a marketing document.

Second, the Plan is

a roadmap.

As a marketing document, the Project Plan must sell the project to the stakeholder(s)

and/or client.

Thus, the Plan should emphasis the emphasis of doing the project.

As a proposal or

“roadmap,” the Project Plan should outline in some detail the likely direction of the project.

This

might include identifying the key variables and methods of analysis.

2

Key Sections of the Project Plan

More specifically, you might consider including the following sections in your Plan.

1. Background statement

2. Conceptual development

3. Variable selection

4. Methods of analysis/analysis plan

5. Form of the results

6. Next steps

In the background statement (section 1), you may wish to sketch out the initial motivation for the

study.

This might include reference to the key stakeholder(s) and/or client.

I recommend targeting

the proposal at a (hypothetical) client to bring a degree of realism to project and to help focus the

project (for example, you could contextualise the study with reference to an Australian bank).

In this

section, also make sure to sell the project.

What are the likely benefits of doing the project, what

new insights do you anticipate and how will these improve decision making for example?

You might find value in a section 2 that outlines the conceptual framework for your project work.

If

you focus your project on customer engagement with banks, for example, you might give some

thought to advantages to banks and their customers from greater engagement and the process that

might drive customers to respond favourably to a bank’s marketing efforts.

My preference is you

use your own common sense and logic to define the key concepts and to develop a rationale for

their links.

I do not expect a review of the literature, but you might find some desk (Google)

research helpful in identifying past studies that have explored similar issues to the ones you are.

A

boxes and arrows diagram might help to illustrate the core concepts and relationships.

The section on variable selection is probably the key section (section 3).

Be very specific about the

variables you intend to study.

In the social science tradition, much emphasis is placed on explaining

why the variables selected for study have been selected – the focus is explanation rather than

prediction.

This is less the case with the data science paradigm with its focus on prediction –

business analysts/data scientists may wish to specific a (initial) model that includes all of the possible

feature variables.

My minimum expectation for this section is that you provide some description of

the output and feature variables you intend to study, and why these feature variables.

Section 4 outlines the methods of analysis.

Here I would you to be specific about the models you

might use to analyse the data.

You may have completed the course BSAN2204 Methods of Business

Analytics.

A focus of that course was predicting a numeric output variable (“song hotness”) using

linear regression.

For this course (BSAN2205 Machine Learning for Business), our target variable is

categorical: it records whether customers opened or did not open a new account in response to the

Bank’s marketing efforts.

My expectations for section 4 are that you can identify an appropriate

statistical model(s) for analysing the data, state something about the assumptions of the model, and

perhaps list the key steps in employing the model.

You could also write out the specific model you

intend estimating (write out the regression equation, for example, with reference to the y- and x- variables).

Section 5 – form of the results – should give an indication of what the outputs might look like.

You

could do mock-up of the results.

You could also say that you will document the results in

PowerPoint format and present them verbally.

The next steps section concludes the proposal.

Here

you might remind the client of the core benefits and indicate you need to initialise the project (final

client sign-off, for example).

You could also add a timeline or perhaps Gantt chart (timetabling the

key activities, when you will do them, and identifying any critical paths).

At this stage, refrain from

3

doing any statistical analysis of the data – save the analysis for the project reports.

Use the Plan to

develop some general knowledge of the models you intend to use and sketch out your best plan for

the analysis you intend to implement.

The final section of your Plan might address next steps (Section 6). You can briefly restate the main

motivation for your Plan and highlight the key “next steps.” Remember the Plan is a marketing

document – perhaps remind the reader of the Plan that this project is an important one and should

be completed now.

The Bank Marketing Dataset

The project work for this Semester uses the Bank Marketing dataset.

Several variations of the

dataset exist.

There is one variation available from the UCI Machine Learning Repository and

another variation on Kaggle.

We will use the version of the dataset available from Kaggle (with some

minor variations).

Owned by Google, Kaggle is an online community of business analysts and data

scientists.

Users can freely upload and download data to and from the site (kaggle.com).

Kaggle

runs competitions often sponsored by third parties.

I encourage you to explore the Kaggle website

and join the Kaggle community.

Kaggle is a great place for those with an interest in machine

learning.

I have downloaded the dataset from Kaggle, introduced some further variations, and placed the

dataset to the Blackboard site.

Please use this version of the dataset for your project.

Appendix A

provides a list of the variables in the Bank Marketing dataset, including brief descriptions.

The target

or output variable is customers’ responses to a recent marketing campaign run by the Bank (the

Bank being a European bank, specifically, a Portuguese bank).

The data is real-world data offered

freely by the Bank to the data science community.

The data consists of 21 variables (the target

variable and 20 feature variables) and observations on approximately 40,000 customers targeted

with a particular marketing campaign.

The output variable is a binary categorical variable –

customers responded to the marketing campaign by either opening a new account or not.

The 20

feature variables include a mix of variables reflecting customers’ characteristics (age, education,

etc.), the nature and status of their existing accounts with the Bank (type of accounts, accounts in

debit, etc.), variables describing the campaign (number of customer contacts during the campaign),

and socio-economic variables (consumer confidence, etc.).

The feature variables are a mix of

categorical and numeric variables.

Given the output variable is a (binary) categorical variable you should explore model forms other

than linear regression.

As a starting point, I recommend you fit a logistic regression model to the

data and subsequently use tree-based methods.

A comparison of these methods could be an

important of your overall project (logistic regression vs decision trees).

Further, you might explore

ensemble methods to enhance your implementation of tree-based methods.

We will cover these

methods in the coming weeks!

Submission Guidelines

The Project Plan has a weight of 20 percent of your score for the course.

Please submit your Plan in

the form of a written Word document.

I expect you could easily write 2,000 words.

Try not to write

more than 3,000.

I will give your Plan a score out of 100.

I will also provide you with written

feedback.

When marking the Project Plan, I will be looking closely at the links between the sections

as much as what you write in each individual section.

For example, the background statement

should set-up the conceptual development that in turn should set-up the variable selection etc.

A

high scoring Plan will have a degree of novelty to it (a unique and/or compelling contextualisation, a

4

thoughtfully specified analysis plan – including appropriate performance metrics, etc.).

Finally, these

notes are a guide only to preparing your Project Plan.

You may find other ways to present it that are

more compelling, more compact, and more complete.

If in doubt, do what you think is best.

I will separately provide you with the marking criteria for the Project Plan.

Note they will closely

follow the criteria of the Project Plan for the course BSAN2204 Methods of Business Analytics.

5

Appendix A

The Bank Marketing Dataset is based on the “Bank Marketing” UCI dataset, with some variations.

Table 1 below lists the variables in the dataset and offers brief descriptions.

Table A1

Variables and Variable Descriptions

Variable Variable Name Variable Type Units/Category Labels

Age

age Numeric Years

Type of Job job Categorical admin

blue-collar

entrepreneur

housemaid

management

retired

self-employed

services

student

technician

unemployed

unknown

Marital Status marital Categorical divorced

married

single

unknown

Education History education Categorical basic4y

basic6y

basic9y

highschool

illiterate

professionalcourse

universitydegree

unknown

Credit in Default default Categorical no

yes

unknown

Housing Loan housing Categorical no

yes

unknown

Personal Loan loan Categorical no

yes

unknown

Contact Type contact Categorical cellular

telephone

Month of Last Contact month Categorical jan

feb

mar

.

.

.

6

Table A1 (Cont’d)

Variables and Variable Descriptions

Variable Variable Name Variable Type Units/Category Labels

Day of Last Contact day_of_week Categorical mon

tue

wed

thu

fri

Duration of Last Call

duration Numeric Seconds

Number of Contacts

campaign Numeric Counts

Days since Last

Contact

pdays Numeric Days

Prior Contacts

previous Numeric Counts

Response to Last

Campaign

poutcome Categorical failure

nonexistent

success

Cyclical Employment

Variation

emp_var_rate Numeric Index

Consumer Price Index

cons_price_idx Numeric Index

Consumer Confidence

cons_conf_idx Numeric Index

Euro Interbank

Offered Rate

(Euriobor)

euribor3m Numeric Interest rate

Employment Rate nr_employed Numeric Index

Customer Response response Categorical no

yes

51作业君版权所有