Decision Sciences Journal of Innovative Education

Volume 18 Number 4 TEACHING BRIEF

October 2020

© 2020 Decision Sciences Institute

Applying the CRISP-DM Framework

for Teaching Business Analytics
Sanjiv Jaggia†
Orfalea College of Business, California Polytechnic State University, San Luis Obispo,
CA, 93407, e-mail: [email protected]

Alison Kelly
Suffolk University, Boston, MA, 02108, e-mail: [email protected]

Kevin Lertwachara and Leida Chen

Orfalea College of Business, California Polytechnic State University, San Luis Obispo,
CA, 93407, e-mail: [email protected], [email protected]

ABSTRACT
Experiential learning opportunities have been proven effective in teaching applied
and complex subjects such as business analytics. Current business analytics
pedagogy tends to focus heavily on the modeling phase with students often lacking a
comprehensive understanding of the entire analytics process including dealing with
real-life data that are not necessarily "clean" and/or small. Similarly, the emphasis
on analytical rigor of ten comes at the expense of storytelling, which is among the
most important aspects of business analytics. In this article, we demonstrate how the
philosophy of the Cross In dustry Standard Process for Data Mining (CRISP-DM)
framework can be infused into the teaching of business analytics through a term-
long project that simulates the real world analytics process. The project focuses on
problem formulation, data wrangling, modeling, performance evaluation, and
storytelling, using real data and the program ming language R for illustration. We
also discuss the pedagogical theories and tech niques involved in the application of
the CRISP-DM framework. Finally, we document how the CRISP-DM framework
has proved to be effective in helping students navigate through complex analytics
issues by offering a structured approach to solving real-world problems.
Subject Areas: analytics project, business analytics, CRISP-DM, data
wran gling, experiential learning, storytelling, R.

INTRODUCTION
The importance of incorporating business analytics in pedagogy has been well
doc umented (Asamoah et al., 2017; Henke et al., 2016). This trend is further
evidenced by the proliferation of business analytics courses and programs
across universities and by the increasing industry demand for analytics
professionals. Although there

†
Corresponding Author. [email protected]
 612
Jaggia, Kelly, Lertwachara, and Chen 613

are no universally agreed-upon definitions of the term "business analytics," we

fol low the lead supplied by the Institute for Operations Research and the
Management Sciences (INFORMS) to define the term as “the scientific
process of transforming data into insights for the purpose of making better
decisions” (2019). Other schol ars have also provided relevant interpretations
of what the analytics process entails. For example, Wilder and Ozgur (2015)
define business analytics as “the applica tion of processes and techniques that
transform raw data into meaningful infor mation to improve [business]
decision making.” Many business enterprises now describe themselves as
“analytics-based firms” and have become heavily depen dent on data-driven
decision making to improve their organizational performance (Watson, 2013).
As a result, business analysts in these organizations are expected to be fully
knowledgeable and well versed in analytics concepts and techniques.
The CRISP-DM (CRoss Industry Standard Process for Data Mining)
frame work is widely regarded as the most relevant and comprehensive
guiding principle for carrying out analytics projects (Abbasi et al., 2016). In
introducing the CRISP DM framework, Wirth and Hipp (2000) describe
business analytics as a creative process that requires a standard approach to
“help translate business problems into data [analysis] tasks, suggest
appropriate data transformations and data [analy sis] techniques, and provide
means for evaluating the effectiveness of the results and documenting the
experience.” This philosophy has been adopted by business analysts and
practitioners in many industry segments, regardless of the analysis techniques
or computing technologies used in the project.
Despite the wide acceptance and adoption of the CRISP-DM framework
by practitioners, current business analytics pedagogy fails to provide a holistic
ap proach to analytics. Students often find themselves not adequately trained to
deal with real-life projects and the emphasis on analytical rigor often comes at
the ex pense of storytelling, which is among the most important aspects of
business an alytics (Dykes, 2016). As suggested by Heim et al. (2005), student
learning in technology-based disciplines, such as business analytics, can be
enhanced through experiential projects that simulate real-life activities. The
current teaching brief in fuses the philosophy of the CRISP-DM framework
into the teaching of business analytics through a term-long project that
simulates the analytics process. Using real-life data that are not necessarily
"clean" and/or small, we demonstrate how instructors can apply the six phases
of the CRISP-DM framework in hands-on an alytics activities.
Benefits of experiential learning pedagogy have been well documented.
For example, Burch et al. (2019) examine over 89 research studies over a 43-
year span and show that students experience “superior learning outcomes”
when experien tial learning is used. Moreover, as an indicator of its efficacy in
enhancing learning outcomes, experiential learning has been widely
implemented at many universi ties (see, for example, Cardozo et al. (2002) and
Silvester et al. (2002)). For the application in this article, we use the
programming language R for illustration, but all of the analytics tasks can be
similarly completed with any other software such as Python or Analytic Solver
(formerly called XLMiner). As business analytics is considered a creative
process, we argue that the pedagogical focus for teaching this subject should
be placed on problem formulation, data wrangling, modeling, performance
evaluation, and storytelling.
614 Teaching Business Analytics

THE CRISP-DM FRAMEWORK

The CRISP-DM framework was developed in the 1990s by a group of five
com panies: SPSS, TeraData, Daimler AG, NCR, and OHRA (Wirth & Hipp,
2000). It consists of six major phases: business understanding, data
understanding, data preparation, modeling, evaluation, and deployment. The
six phases can be summa rized as follows:
√
Business understanding: According to Wirth & Hipp (2000), this first
phase focuses on “understanding the project objectives and
requirements from a business perspective, and then converting this
knowledge into a data mining problem definition, and a preliminary
project plan designed to achieve the objectives.” Students should be
reminded that this is a crit ical step where business objectives are
identified in order to steer the sub sequent direction of the project.
√
Data understanding: This phase involves an initial data collection and
pro ceeds to activities that help students become more familiar with
the data. Students also need to identify potential data quality
problems, preliminary insights into the data, and possible subsets of
data to form hypotheses that can uncover hidden information.
√
Data preparation: Specific tasks in this phase include data reduction,
data wrangling and cleansing, and data transformation (e.g., creating
dummy variables) for subsequent analyses and testing.
√
Modeling: This phase involves the selection and development of
analytics techniques and models. In addition, portions of a data set
are often set aside for training and validating the model(s).
√
Evaluation: This phase involves reviewing and interpreting the
analysis results in the context of the business objectives and success
criteria de scribed in the first phase.
√
Deployment: During this final phase, the knowledge gained from data
analysis is translated into a set of actionable recommendations. In
addi tion to performing appropriate analysis, analysts need to
understand that effectively communicating the analysis results to
business constituents also plays a key role in a successful analytics
project.
The CRISP-DM framework implies a cyclical nature of business analyt
ics projects, and therefore, is often depicted as a life cycle model as shown in
Figure 1.
In addition to being implemented in industry projects, the CRISP-DM
framework has also been used as a guiding principle in curriculum develop
ment in higher education. For example, in 2018, the University of Chicago
launched a data analytics program whose core courses are “structured along
the CRISP-DM methodology” (2019). Other programs in analytics (see, for
example, Northwestern University’s Master’s in Data Science program, 2019)
often incor porate the CRISP-DM methodology in their curriculum. However,
for individ ual analytics courses, the pedagogical focus is usually on the
modeling phase, which is only one of the six phases in the CRISP-DM
framework (Rudin, 2012).
Jaggia, Kelly, Lertwachara, and Chen 615 Figure 1: CRISP-DM life cycle (Wirth &

Hipp, 2000).
 Students often fail to realize the interplays between the phases and how they
col lectively contribute to the success of analytics projects. Moreover, most
data sets introduced in these courses tend to be small and "clean" with little, if
any, data wrangling required. As such, the data understanding phase and the
data prepara tion phase receive minimal coverage. Finally, as a technical field,
analytics ed ucation tends to overlook the importance of storytelling, the
process of trans lating data points, analytical methodologies, and findings into
interesting stories that are more palatable to the intended audience. The project
described in this teaching brief aims to address these deficiencies witnessed in
the current analytics education.

THE PROJECT
The project can be assigned to teams of students enrolled in a business
analytics course in either the upper division undergraduate curriculum or at the
graduate level. Unlike the pedagogical approach used by Anderson and
Williams (2019)
616 Teaching Business Analytics

Table 1: CRISP-DM phases and corresponding project learning

objectives. CRISP-DM Phases Project Learning Objectives
Business Formulate business questions that lead to
Understanding business strategies or actions.
Data Understanding Describe the data in terms of the business context. Data
Preparation Perform data wrangling to prepare the data for subsequent analyses.
Modeling Develop predictive model(s) to inform decision making. Evaluation
Evaluate model performance and select the best predictive model(s).
Deployment Communicate key findings through storytelling.
where students work on personal analytics topics, such as learning foreign lan
guages, stress management, and personal wellbeing, the current research
focuses on the business processes integrated in the CRISP-DM framework.
Our approach offers students a culminating experience that involves a practical
business topic with project requirements that permeate throughout the CRISP-
DM process. Stu dents are advised to work on the project throughout the term
as each of the six CRISP-DM phases is discussed. The instructor will conduct
periodic reviews of the project progress and provide feedback to student teams.
We provide a sample of the project assignment in Appendix A.
The learning objectives of this project align with the CRISP-DM
framework and are shown in Table 1. In the first phase, students formulate
business ques tions that will ultimately lead to business strategies or actions.
They then describe the data in terms of the business context and perform data
wrangling to prepare the data for subsequent analyses. Following data
description and data wrangling, students apply modeling techniques to the final
data set(s) to produce a number of predictive models that best address the
business questions. They then evaluate model performance to select the best
predictive model(s). Finally, as the deploy ment of the predictive model(s) is
not practical in an educational setting, students focus on communicating key
findings of the project through storytelling in the last phase.

The Data
The data used in this project are the ERIM data set provided by the James M.
Kilts Center of the University of Chicago’s Booth School of Business
(https: //www.chicagobooth.edu/research/kilts/datasets/erim). The ERIM data
set con tains demographic information on 3,189 households in two midwestern
cities in the United States and their purchases in several product categories
(e.g., frozen dinners, yogurt, ketchup, margarine) from participating stores
over a 3-year period (1985-1988). For demonstration purposes, the application
used in this teaching brief focuses only on yogurt and frozen dinners. The
instructor may determine the scope of the project that best aligns with the
objectives of the course. One approach is to ask each team to select a product
category to analyze, and another approach
Jaggia, Kelly, Lertwachara, and Chen 617

is to design the project as a competition where student teams all focus on the
same product category or categories.
It is worth noting that while the project described in this teaching brief
was designed around an existing data set, real-life business analytics projects
would likely start with business managers identifying problems that require
data-driven solutions instead of asking what questions they can answer with
the existing data. The students must understand that the identified business
questions should drive the entire analytics process including the acquisition of
relevant data. The instructor should explain this important limitation of the
project to the students in order to provide them with a realistic expectation of
what they are likely to encounter in real projects.

Business Understanding
The first phase of the project deals with formulating business questions
through an understanding of the business context. As most students are
familiar with the retail industry, students can identify the potential business
opportunities the data set presents to retailers and manufacturers in marketing
and selling these products. For consistency across student teams, we suggest
that the following two business questions be included in the assignment:
1. Which households are likely to purchase yogurt and frozen dinner
prod ucts?
 2. How much money is each household likely to spend on each product?
Students also need to understand that in practice, documenting available
re sources, estimating potential costs and benefits of the project, and
identifying suc cess criteria are also part of this first phase. In addition,
business analysts often work with a wide range of stakeholders, and therefore,
identifying relevant stake holders and understanding their values and
requirements are critical to the success of the project. During this phase of the
project, the instructor may also impress upon students the general differences
between supervised and unsupervised techniques. For example, students may
be asked to consider whether supervised (predictive modeling) or unsupervised
(pattern recognition) learning would be appropriate for achieving the analytical
objectives and whether the current data set supports these techniques. These
questions can be further explored during the data understanding phase of the
CRISP-DM framework.

Data Understanding
Depending on the prerequisite knowledge of the students, the instructor can
choose to require students to download the original data from the ERIM
website, which require a fair amount of data integration and preprocessing.
The original household data set contains 3,189 observations with 62 variables.
Given that we are interested in variables that may influence a household’s
decision to purchase yogurt or frozen dinners, we remove irrelevant variables
from the data set such as the head of household’s first name and whether the
household had a washer and dryer. Also, because nearly all male and female
households are white in this data
618 Teaching Business Analytics

Table 2: Summary of yogurt and dinner expenditures.

Expenditure Min 1st Quartile Median Mean 3rd Quartile Max

Yogurt 0.00 1.20 10.33 40.60 35.84 3,258.40 Frozen Dinners 0.00 0.00 0.00 55.77
10.45 4,073.01

set, we delete the race variables. After integrating the household data with
detailed purchase records, we produce the modified data set, now renamed
ERIMData, which contains 3,189 observations and 18 variables. We provide a
complete list and description of the variables in Table B1 in Appendix B; all of
the data used in this application are available upon request.
Even with a more manageable data set, data preparation and wrangling
are still necessary prior to model development and analysis. In this application,
data wrangling and analytical modeling are performed using the R language;
however, these are common tasks that can also be completed with other
software packages or programming languages. For those who wish to learn R,
basic tutorials can be found at www.r-project.org/about.html and
www.rstudio.com/online-learning. During this phase, students also explore the
data and identify possible variables that may add value to subsequent analysis
phases. In Appendix C, we provide a portion of the R code used for data
wrangling and selected business analytic models. The complete R code is
available upon request.
It is a common practice to produce summary statistics, look for
symmetry, and/or identify outliers for key variables. Table 2 provides a
summary for the two expenditure (target) variables. Even with only descriptive
statistics, students can draw insights from the data. For both target variables,
the median is notably less than the mean and the maximum value is
dramatically higher than the third quar tile. Thus, it is likely that both
distributions are positively skewed and have outliers. Students are encouraged
to use data visualization, such as boxplots, to reinforce this finding, and to
explore other visualization tools, such as histograms, stacked column charts,
scatterplots, bubble plots, and heat maps, to discover other interest ing patterns
and stories.
The strong evidence of positive skewness and/or outliers suggests the
follow ing two approaches for conducting predictive analytics:
1. Log-transform the yogurt and dinner expenditure variables for
prediction models.
2. Bin the yogurt and dinner expenditure variables for classification models.
These transformations along with other potential data-related issues can
then be dealt with in the data preparation phase of the CRISP-DM framework.

Data Preparation
Although the ERIM data set is from the 1980s, it highlights many of the
relevant issues that business analysts still confront daily, starting with
transforming un wieldy raw data into useful, actionable information. During
the data preparation stage, we ask students to develop a comprehensive plan
for data wrangling,
Jaggia, Kelly, Lertwachara, and Chen 619

which is the process of cleansing, integrating, transforming, and enriching the

data. Most analysts will attest that data wrangling is one of the most critical
and time-consuming steps in any analytics project.
With 3,189 observations and 18 variables in ERIMData and preliminary
ideas for potential analysis techniques, students are advised that further data
preparation is still necessary. In many data sets, including this one, there are
missing values. For this project, it is appropriate to replace all missing values
with 0s. For example, missing information on male head of household implies
that there is no male head of household. Similarly, missing information on
yogurt expenditure implies that there is no expenditure on yogurt. The
instructor, however, should remind students that it is not always appropriate to
replace missing values with 0s.
Next, students are told that data wrangling is needed to create/transform
vari ables that make good economic sense for the analysis. We create 12
additional vari ables in this phase. For instance, we create a Yogurt variable
that assumes the value of 1 if a household purchases yogurt over the time
period, 0 otherwise. Similarly, we create a Dinner variable that assumes the
value of 1 if a household purchases frozen dinners over the time period, 0
otherwise. With the 12 new variables, we no longer need 13 of the original
variables, and therefore, we remove them to create the final data set for
analysis with 3,189 observations and 17 variables (18 + 12 – 13 = 17). We
provide a complete list and description of the newly created variables as well
as the ones retained from the original data set in Table B2 of Appendix B.
It is always useful to produce descriptive statistics on key variables in
the final data set, with a focus on the business questions previously discussed.
One possible project requirement is to have students create subsets of the data
based on whether or not the households purchased the two products of interest:
yogurt and frozen dinners. Table 3 shows averages for potential predictor
variables for the entire sample and for households in each of subsets.
Students are encouraged to grasp key characteristics of the data. They
can also draw some interesting observations from subsets of the data. For
example, on average, households that purchased yogurt earned more, worked
longer hours, and had a younger head of household as compared to households
who did not purchase yogurt. Similar observations can be drawn when
comparing households that did and did not purchase frozen dinners.
 Modeling
The modeling process involves applying predictive models to the data to iden
tify hidden structures, patterns, and relationships among variables. The
efficacy of competing models is assessed in this phase as well. Therefore, we
tend to divide this phase into two subphases, model development and model
assessment.

Model Development
In the model development subphase, the instructor should highlight the
strengths and limitations of various modeling techniques. Students will
identify and choose the appropriate analytical techniques after considering
their advantages and limi tations. Initially, students may start with a model that
offers a higher level of in terpretability. For example, both logistic regression
for classification and multiple
620 Teaching Business Analytics

Table 3: Sample averages in all households (HH) and in subsets. Yogurt

Frozen dinners

Variable All Purchase No purchase Purchase No purchase

HH Income 6.0066 6.1805 5.2407 6.0668 5.9765 Cable 0.6579 0.6637 0.6322
0.6651 0.6543
Single Family Home 0.8736 0.8823 0.8356 0.9012 0.8598

Own Home 0.8498 0.8619 0.7966 0.8702 0.8396 Pets 0.5165 0.5321 0.4475 0.5823
0.4835 Married 0.7294 0.7430 0.6695 0.7883 0.6999
College Educated Both 0.1154 0.1301 0.0508 0.0988 0.1237 0.0618
College Educated One
0.0635 0.0542 0.0442 0.0706

Work Hours 27.4955 28.4386 23.3407 29.2324 26.6270

Other HH Members 1.0019 1.0843 0.6390 1.2775 0.8641

Age 48.2469 47.2564 52.6102 44.7662 49.9873 Female HH 0.2405 0.2336 0.2712
0.1910 0.2653

Number of HH 3189 2599 590 1063 2126

linear regression for prediction are highly interpretable and quantify the impact
of each predictor variable on the target variable.
Table 4 shows the logistic regression results for classifying whether or
not a household purchases yogurt and frozen dinner products, respectively. As
part of the project assignments, the instructor should ask students to consider
the following questions based on the initial modeling results: (a) which
predictor variables are the most influential predictors?, (b) how much impact
does each predictor variable have on the probability of a household purchasing
yogurt/dinner?, and (c) which type of household is likely to purchase
yogurt/dinner? The instructor may also ask students to compare answers to
these questions to their initial assumptions gained from data exploration during
the earlier phases.
Students are also asked to estimate multiple linear regression models on
the expenditure variables. As mentioned before, given the skewness of the
numeri cal target variables, these target variables must first be transformed into
natural logs. The instructor should provide the correct interpretation of the
estimated co efficients of log-linear models. Qualitatively, the results from the
multiple linear regression models are similar to those of the logistic models;
the results are not reported in the article for the sake of brevity.
While logistic and linear regression models offer important insights on
how the predictor variables influence the target variables, the instructor can
remind stu dents that predictive modeling focuses more on the model’s ability
to classify or predict a future case correctly than trying to interpret or draw
inferences from the model. In other words, a well-performing explanatory
model may not necessarily be a good predictive model. Data-driven
techniques, such as naïve Bayes, ensemble
Jaggia, Kelly, Lertwachara, and Chen 621

Table 4: Estimated logistic regression models for yogurt and dinner. Variable
Yogurt Frozen dinners
Intercept 0.3047 (0.447) (0.349)
HH Income 0.0582 ** (0.007) 0.1367
Cable –0.0052 (0.959) (0.365)
Single Family Home –0.0614 (0.714) 0.1359
Own Home 0.3565 ** (0.022) (0.326)
Pets 0.0740 (0.456) 0.2186 ** (0.007)
Married 0.6901 ** (0.004) 0.4194
College Educated Both 0.6808 ** (0.001) (0.109)
College Educated One 0.3977 * (0.073) –0.4930 ** (0.000)
Work Hours 0.0014 (0.717) –0.2248
Other HH Members 0.2455 ** (0.000) (0.250)
Age –0.0095 ** (0.044) –0.0045
Female HH 0.9031** (0.000) (0.165)
–0.0870 0.1735 ** (0.000)
(0.814) –0.0214 ** (0.000)
–0.0387 ** (0.025) 0.1424
–0.0784 (0.585)
Notes. Parameter estimates with the p-values in parentheses; * and ** represent
significance at the 10% and 5% level, respectively.

trees, and k-nearest neighbors, may result in better predictive models even
though they suffer in interpretability.

Model Assessment
In the model assessment subphase, the instructor should stress the importance
of evaluating model performance using the validation or test data set instead of
the training set. Performance measures should evaluate how well an estimated
model will perform in an unseen sample, rather than making the evaluation
solely on the basis of the sample data used to build the model. The validation
data set not only provides measures for evaluating model performance in an
unbiased manner but also helps optimize the complexity of predictive models.
Students are asked to evaluate model performance using the validation
data set each time a model is developed and focus on measures of predictive
perfor mance rather than on goodness-of-fit statistics as in a traditional
analytical process. For classification models, performance measures include
the accuracy rate, sensi tivity, and specificity. For prediction models, common
performance measures are
622 Teaching Business Analytics

Table 5: Performance measures of logistic models.

Yogurt Frozen dinners

Measure Cutoff = 0.5 Cutoff = 0.82 Cutoff = 0.5 Cutoff = 0.33

Accuracy 0.8141 0.5890 0.6541 0.5890 Sensitivity 0.9971 0.5833 0.1106 0.5671
Specificity 0.0085 0.6144 0.9259 0.6000

the mean error (ME), the root mean square error (RMSE), and the mean
absolute error (MAE). Furthermore, performance charts, such as the
cumulative lift chart, the decile-wise lift chart, and the receiver operating
characteristic (ROC) curve are also used to evaluate model performance.
To illustrate the teaching points, we partitioned the data to re-estimate
and assess the logistic regression model for classifying whether a household
would purchase yogurt or frozen dinners, respectively; see Table 5 for the
performance measures. We present two sets of performance measures, using
the cutoff value of 0.5 (the default cutoff value for binary classification
models) and the cutoff value equal to 0.82 for yogurt and 0.33 for frozen
dinners (the actual proportion of households in the data that purchased yogurt
and frozen dinners, respectively). It is important to point out to students that
classification performance measures are highly sensitive to the cutoff values
used. A higher cutoff value classifies fewer number of cases into the target
class, whereas a lower cutoff value classifies more cases into the target class.
As a result, the choice of the cutoff value can influence the confusion matrix
and the resulting performance measures. In cases where there are asymmetric
misclassification costs or an uneven class distribution in the data, it is
recommended that the proportion of target class cases be used as the cutoff
value. For example, by setting the cutoff value to 0.33 for frozen dinners, the
model generates a sensitivity value of 0.5671 meaning that 56.71% of the
target class cases are correctly classified, versus a sensitivity value of 0.1106 if
the cutoff value is 0.5.
It is sometimes more informative to have graphical representations to
assess model performance. Figure 2 displays these charts that are associated
with the lo gistic regression model for frozen dinners. Note that the charts are
created using the validation data set. Unlike the numeric performance
measures, the performance charts are not sensitive to the choice of cutoff
value. Students need to be able to articulate the performance of various models
based on the performance charts. For example, Figure 2 suggests that the
logistic regression model offers improvement in prediction accuracy over the
baseline model (random classifier). The lift curve lies above the diagonal line
suggesting that the model is able to identify a larger per centage of target class
cases (households that purchase frozen dinners) by looking at a smaller
percentage of the validation cases with the highest predicted proba bilities of
belonging to the target class. The decile-wise lift chart conveys similar
information but presents the information in 10 equal-sized intervals. Finally,
the ROC curve also suggests that the model performs better than the baseline
model in terms of sensitivity and specificity across all possible cutoff values.
The area
Jaggia, Kelly, Lertwachara, and Chen 623 Figure 2: Performance charts for logistic

classification for frozen dinners.

 under the curve (AUC) value is 0.6138, which is larger than the AUC value of
the baseline model (AUC = 0.5).
In the modeling phase of the CRISP-DM framework, students were
asked to consider classification and prediction models. We show the logistic
regression model for classification and the multiple linear regression model for
prediction. While the logistic model seemed to work well for both products in
this applica tion, students should verify that other data-driven techniques, such
as ensemble trees and k-nearest neighbors, do not yield better performance
measures. As an alternative to the multiple linear regression model, students
may want to consider a regression tree. Figure 3 displays the regression tree
model and its performance measures for frozen dinners, which can be used as
the basis for model selection. For the regression tree, the amount a household
spends on frozen dinners is de termined by the number of members in the
household and age. Based on RMSE, this model has a smaller average
prediction error than the multiple linear regres sion model. Students are
encouraged to verify this fact and contemplate why a simpler model may
produce more accurate predictions than the more complex ones do in some
cases. While we focus on supervised learning in this application, students can
also explore unsupervised learning approaches (e.g., cluster analy sis) to
further examine interesting patterns and relationships that may exist in the
data.

Evaluation
While the efficacy of the predictive models with regard to various performance
measures is assessed during the modeling phase, the evaluation phase focuses
on determining whether the models have properly achieved the business
objectives specified in the earlier phases. This phase reminds students that data
mining is not merely an academic exercise but a field designed to impact
decision making, and it requires students to take off the hat of a technical
expert and put on the business hat.
624 Teaching Business Analytics Figure 3: A regression tree model and its

performance measures.
One important process within the evaluation phase is to review the steps
executed to construct the model to ensure that no important business issues
were overlooked. Therefore, each team selects two students outside the team to
review and critique the team’s modeling process and validate the logic behind
the process.
This phase also impresses on students the importance of domain
knowledge and business acumen in understanding the findings of analytics.
During this phase, students frequently realize that the strongest patterns and
relationships identified by the models are often obvious, less useful, or simply
reflect business rules, and in many cases, the most technically elegant or
sophisticated solutions yield little relevant insights to answer the business
questions. In addition to self-assessment, student teams are asked to
collaborate with domain experts to evaluate how well their predictive models
achieve the business objectives. In the case of the ERIM project, each team
conducts interviews with the instructor, who plays the role of a retail expert, to
discuss the team’s findings and explore the possible actionable decisions that
retailers and manufacturers can make based on the findings.
For example, the classification models reveal interesting differences and
sim ilarities between households that are likely to purchase yogurt and those
that are likely to purchase frozen dinners. Households that are likely to
purchase yogurt tend to have higher income and education levels and consist
of a married couple or are led by a female head of household; whereas
households that are likely to
Jaggia, Kelly, Lertwachara, and Chen 625

purchase frozen dinners tend to have lower income and education levels and
have at least one pet. Relatively young head(s) of households with large
families are expected to purchase both yogurt and frozen dinners. Students are
encouraged to develop compelling data stories that help depict the profiles of
these households for the audience and provide actionable recommendations
that would lead to market ing and advertising, store placement, and product
 design strategies. Such discus sion often leads to teams backtracking to earlier
phases to augment data preparation and modeling processes.
Putting on the business hat encourages students to look at the models
from a different perspective. For example, while prediction models produce
predicted values of the target variable, a marketing executive may decide to
place more em phasis on the ranking of the predicted values rather than the
values themselves. Similarly, in order to achieve our business objective, we are
more likely to be in terested in identifying households that would spend more
on frozen dinners so that we can target these households for future marketing
efforts rather than accurately predicting how much each household spends on
frozen dinners. The performance measures, such as RMSE, often fail to
provide us with this critical information. To understand the model’s ability to
correctly rank spending, performance charts, such as the lift chart and the
decile-wise lift chart, can be more helpful. A criti cal evaluation of the
analytics findings from the business perspective in this phase helps the teams
refocus on the objectives of the project and create compelling data stories and
recommendations for business decision makers.

Deployment
The final phase of the project involves a written report and presentation of the
key findings. This phase stresses the importance of storytelling to
communicate ana lytical findings to their intended audience effectively.
Storytelling, or data story telling, refers to crafting and delivering compelling
data-driven stories to decision makers for the purpose of converting insights
into actions; that is the final phase of the CRISP-DM framework.
Contrary to popular belief, storytelling is not the same as data
visualization although presenting data through visually engaging figures and
diagrams is a part of storytelling. Students are asked to focus on three key
elements of storytelling: data, visualization and narrative, and how they
complement one another to create a compelling story about the findings.
Simply presenting the analytical process and findings from a technical
perspective would have limited use to decision makers. To engage the
audience, students must learn to focus on the context around the data that helps
demonstrate the business value of the analysis and use appropriate and
engaging visualizations to help reveal the underlying patterns and relation
ships. Students are advised to present business insights gained from data
analysis from a nontechnical standpoint and craft the story around the data by
focusing on answering the following three questions:

(1) Why should the decision maker care about the findings?
(2) How do these findings affect the business?
(3) What actions do you recommend to the decision maker?
626 Teaching Business Analytics

Storytelling gives the dry topic of data analysis an interesting spin and
makes the content of the report and presentation more palatable to the audience
who are often business decision makers with little training and/or interest in
analytical methodologies. We find that storytelling is often intimidating at first
to students in the business analytics course, especially to those with a technical
mindset. How ever, it is a career-building skill that can be improved with
practice and guidance from the instructor.
Critical reflection is an essential component of the experiential learning
cycle (Kolb, 2015). It helps enhance students’ understanding of the
experiential activi ties in the context of the learning objectives of the course.
Upon the completion of the project, the students are asked to reflect on their
analytics work. A critical reflection framework, such as the self-reflective
model, by Rolfe et al. (2001) can reinforce students’ learning experience. Their
reflective model is based on three simple steps: "What?," "So what?," and
"Now what?." In the "What?" step, stu dents reflect upon important questions,
such as "What happened in the project?," "What was the role of each team
member?," and "What was the problem being solved?" During the "So what?"
step, students may consider questions, such as "What other issues and
opportunities arose from the project?," "What conclusions did you draw from
the project?," and "What did you learn about the project and other team
members?" Finally, during the "Now what?" step, students contemplate
questions such as "How will you apply what you learned from the project?,"
"If you need to complete a similar project again, what would you do
differently?" and "What other skills might be beneficial to learn before you
proceed to the next project?"

ASSESSMENT OF PEDAGOGICAL EFFECTIVENESS

After completing an analytics course that provided students with a deep dive of
the CRISP-DM framework, groups of graduate-level business analytics
students then enrolled in an industry project course. In this course, students
followed the six CRISP-DM phases to complete an analytic project addressing
real-world business problems using large data sets. This sequence allowed us
to assess the effectiveness of the CRISP-DM framework as a pedagogical
benchmark. Students filled out a course evaluation survey at the end of the 10-
week quarter. The survey results were not released to the instructor until after
the grades were submitted.
Generally, students responded positively to the CRISP-DM framework.
Us ing the course evaluations from the same course that was offered in Spring
2017 where the CRISP-DM framework was not used as the pedagogical
benchmark, we have seen improvements across multiple student satisfaction
and learning measurements. In Winter 2019, approximately 89% of the
students responded to the survey "strongly agreed" or "agreed" that the course
was “educationally effective,” whereas 83% students responded positively to
this question in Spring 2017. In addition, 83% indicated that their interest in
business analytics “has been stimulated” by the project versus 67% of the
students in Spring 2017. Regarding the amount of workload, 89% responded
that the workload was “appropriate in relation to other courses of equal credit”
versus 83% of the students in Spring 2017. The students expressed that the
CRISP-DM framework proved effective in
Jaggia, Kelly, Lertwachara, and Chen 627

helping them navigate through complex analytics issues and offered a

structured approach to solving real-world, big data problems. While the
anecdotal evidence from both students and the instructor regarding the
effectiveness of the CRISP-DM framework was extremely positive, more data
will need to be collected in the future to validate our conclusion empirically.
At the end of the quarter, each student group presented its analysis
results and key findings following the storytelling guidelines outlined in this
teaching brief. The presentations focused on how the project delivered
business value through a structured process of analytics with the intended
audience being business execu tives rather than analytics professionals. Every
student participated in the group presentation and was required to meet
individually with the instructor to discuss his or her participation and
contribution to the project. Students were graded on their mastery of the key
knowledge points throughout the CRISP-DM phases and ability to
communicate the findings effectively. For the Winter 2019 quarter, the oral
presentations and individual meetings showed that approximately 82% of the
students were able to articulate the business value of their projects and commu
nicate the key findings to a nontechnical audience effectively. Almost all of the
students demonstrated a satisfactory level of understanding of the CRISP-DM
 framework.

CONCLUSIONS
Experiential learning opportunities that mimic real-world projects have been
proven effective in teaching applied subjects such as business analytics. The
project described in this teaching brief provides students with a holistic
experience of converting data into insights and actionable business strategies.
The infusion of the CRISP-DM framework throughout the project creates a
structured approach to a creative problem-solving process. This extensive
project is valued by both the instructor and students. The structure of the
project and instructional experience gained by the instructor from this project
can be readily applied to other large data sets and business problem contexts
including consulting projects. Students who have undertaken this project gain a
better understanding of the CRISP-DM framework, which is a widely adopted
industry standard for analytics projects, and an integrated view of the
knowledge points that permeate throughout the business analytics curriculum.
The project provides an effective and engaging experiential learning activity
that helps improve career readiness for business students.

REFERENCES
Abbasi, A., Sarker, S., & Chiang, R. H. (2016). Big data research in
information systems: Toward an inclusive research agenda. Journal of
the Association for Information Systems, 17(3).
Anderson, J. S. & Williams, S. K. (2019). Turning data into better decision
mak ing: Asking questions, collecting and analyzing data in a personal
analytics project. Decision Sciences Journal of Innovative Education,
17(2), 126–145.
628 Teaching Business Analytics

Asamoah, D. A., Sharda, R., Hassan, Z. A., & Kalgotra, P. (2017). Preparing a
data scientist: A pedagogic experience in designing a big data analytics
course. Decision Sciences Journal of Innovative Education, 15(2), 161–
190.
Burch, G. F., Giambatista, R., Batchelor, J. H., Burch, J. J., Hoover, J. D.,
Heller, N. A. (2019). A meta-analysis of the relationship between
experiential learning and learning outcomes. Decision Sciences Journal
of Innovative Education, 17(3), 239–273.
Cardozo, R. N., Durfee, W. K., Ardichvili, A., Adams, C., Erdman, A. G.,
Hoey, M., Iaizzo, P. A., Mallick, D. N., Bar-Cohen, A., Beachy, R., &
Johnson, A. (2002). Perspective: Experiential education in new product
design and business development. Journal of Product Innovation
Management, 19(1), 4–17.
Dykes, B. (2016). Data storytelling: The essential data science skill everyone
needs. Forbes. https://www.forbes.com/sites/brentdykes/2016/03/31/data
storytelling-the-essential-data-science-skill-everyone-needs/
Heim, G. R., Tease, J., Rowan, J., & Comerford, K. (2005). Experiential
learning in a management information systems course: Simulating IT
consulting and CRM system procurement. Communications of the
Association for Informa tion Systems, 15, 428–463.
Henke, N., Bughin, J., Chui, M., Manyika, J., Saleh, T., Wiseman, B., &
Sethupathy, G. (2016). The age of analytics: Competing in a data-driven
world. McKinsey Global Institute. https://www.mckinsey.com/business
functions/mckinsey-analytics/our-insights/the-age-of-analytics-
 competing in-a-data-driven-world.
Institute for Operations Research and the Management Sciences. (2019). Best
Definition of Analytics. https://www.informs.org/About-
INFORMS/News Room/O.R.-and-Analytics-in-the-News/Best-
definition-of-analytics
Kolb, D. (2015). Experiential learning: experience as the source of learning
and development. New Jersey: Pearson Education, Inc.
Northwestern University. (2019). Course Descriptions and Schedule.
https://sps.northwestern.edu/masters/data-science/program-courses.php?
course_id=4790
Rolfe, G., Freshwater, D., & Jasper, M. (2001). Critical reflection in nursing
and the helping professions: A user’s guide. Basingstoke: Palgrave Macmillan.
Rudin, C. (2012). Teaching ‘Prediction: Machine Learning and Statistics’. Pro
ceedings of the 29th International Conference on Machine Learning, Edin
burgh, Scotland.
Silvester, K. J., Durgee, J. F., McDermott, C. M., & Veryzer, R. W. (2002).
Per spective: Integrated market-immersion approach to teaching new
product de velopment in technologically-oriented teams. Journal of
Product Innovation Management, 19(1), 18–31.
University of Chicago. (2019). Curricular Updates Spring 2018. https://
grahamschool.uchicago.edu/news/curricular-updates-spring-2018
Jaggia, Kelly, Lertwachara, and Chen 629

Watson, H. J. (2013). Business case for analytics. Biz Ed, 49–54. Wilder, C. R.
& Ozgur, C. O. (2015). Business analytics curriculum for undergrad uate
majors. INFORMS Transactions on Education, 15(2), 180–187. Wirth, R. &
Hipp, J. (2000). CRISP-DM: Towards a standard process model for data
mining. Proceedings of the Fourth International Conference on the Practical
Application of Knowledge Discovery and Data Mining, Manch ester, United
Kingdom, 29–39.

APPENDIX A: SAMPLE BUSINESS ANALYTICS

PROJECT ASSIGNMENT
This project is based on the data set called ERIMdata.xlsx that includes
about 3,000 households in two midwestern cities in the United States. The data
contain demographic information such as household incomes, number of
household mem bers, education levels of the heads of households, as well as
information on the purchases of a number of retail products such as frozen
dinners and yogurt. The data were collected between 1985 and 1988 by the
marketing research firm, AC Nielsen.
Your assignment is to first propose a business analytics plan based on
the CRISP-DM framework and identify and complete the appropriate tasks for
each of the six CRISP-DM phases. The project deliverables included in a final
written report and an oral presentation should follow the outline shown below:
Business understanding: Describe the business opportunities that the data
present and formulate relevant business questions.
Data understanding: Explore the data set with descriptive analytics tools and
pro vide relevant information. Examine the possibility of supervised and
unsuper vised analysis techniques and identify possible variables for further
analysis. Keep in mind the business opportunities and questions formulated
in the first phase. The following criteria may also be considered as a guide:

• Does a target variable(s) exist?

 • Does the data set contain historical values of the target variable(s)? •
Does the data set have a sufficient number of observations to support
data partitioning?

Data preparation: Determine and perform the necessary data wrangling and
prepa ration tasks based on the decision made during the business and data
understand ing phases. Explain the rationale for these tasks and document the
changes that you have made to the data set.
Modeling: Consider the strengths and weaknesses of different modeling tech
niques. Implement the appropriate techniques, explain the rationale for your
selections, and present relevant analysis results and interpretation. For the su
pervised techniques, determine whether to use classification or prediction
mod els and explain your decision. Use appropriate data partitioning and
performance measures to evaluate the competing models implemented in the
modeling phase. Identify the best model(s).
630 Teaching Business Analytics

Evaluation: Refocus on the business objectives of the project. Review the steps
ex ecuted to construct the model to ensure no key business issues were
overlooked. Evaluate whether the models have properly achieved the
business objectives out lined during the business understanding phase.
Formulate actionable recommen dations based on the findings.
Deployment: Communicate the findings and relevant business insights with a
written report and oral presentation that incorporate appropriate statisti cal
information and visuals. The main focus should be placed on provid ing
actionable business recommendations for a managerial and non-technical
audience.

APPENDIX B: DATA DICTIONARY

Table B1: Description of variables in ERIMData2.

Variable Description

HH_ID The household’s identification number

ResType Types of residence: 1 for Apartment, 2 for Condo, 3 for Single Family, 4 for
Multiple Family, 5 for Mobile, and 6 for Other.
ResStatus Residence status: 1 for owned home, 2 for rented, and 3 for other. HHInc
The average annual income of a household; there are 14 categories for this variable.
HHNbr The number of members in the household.
MWrkHrs The average hours worked each week by the male head of household. MEdu
Education level of the male head of household: values less than 9 imply varying
education levels prior to a college degree, 9 for graduated from college, 10 for attended
graduate school, and 11 for post-graduate degree. FWrkHrs The average hours worked
each week by the female head of household. FEdu Education level of the female head
of household. See MEdu for detail. FBirth The birth year of the female head of
household.
F_Rel Relationship within the household: 1 for female head of household, 2 for male
head of household, 3 for daughter, 4 for son, and 5 for other.
MBirth The birth year of the male head of household.
M_Rel Relationship within the household: 1 for female head of household, 2 for
male head of household, 3 for daughter, 4 for son, and 5 for other.
Cable Whether or not the household has cable; 1 if yes, 0 otherwise.
Cats Whether or not the household has cats; 1 if yes, 0 otherwise. Dogs
Whether or not the household has dogs; 1 if yes, 0 otherwise. YogExp A
household’s yogurt expenditures (in $)
DinExp A household’s frozen dinner expenditures (in $)
2
Data are available upon request.
 Jaggia, Kelly, Lertwachara, and Chen 631

Table B2: Description of variables used for model development and

analysis3. Variable Description

HH_ID The household’s identification number

Cable Whether or not the household has cable; 1 if yes, 0 otherwise. HHInc The
average annual income of a household; there are 14 categories for this variable.
YogExp A household’s yogurt expenditures (in $)
DinExp A household’s frozen dinner expenditures (in $)
Yogurt Based on YogExp, it assumes the value of 1 if a household purchases yogurt
over the time period, 0 otherwise.
Dinner Based on DinExp, it assumes a value of 1 if a household purchases frozen
dinners over the time period, 0 otherwise.
Sglfam Based on ResType, it assumes the value of 1 if a household resides in a
single-family home, 0 otherwise.
Ownhome Based on ResStatus, it assumes the value of 1 if a household owns a home,
0 otherwise.
Pets Based on Cats and Dogs, it assumes the value of 1 if a household has a cat or a
dog, 0 otherwise.
Married Based on M_Rel and F_Rel, it assumes a value of 1 if a household has both a
male and female head of household, 0 otherwise
EduBoth Based on MEdu and FEdu, it assumes the value of 1 if both heads of
households have at least a college degree, 0 otherwise
EduOne Based on Married, MEdu, and FEdu, it assumes the value of 1 if a
household led by a single person has at least a college
degree, 0
otherwise.
WrkHrs Based on Married, MWrkHrs, and FWrkHrs, it equals half of the hours
worked by the two heads of the household or hours worked by the
single head of the household.
HHMembers Based on Married and HHNbr, it counts the additional members
in a household excluding the head(s) of household.
Age Based on MBirth and FBirth, it equals half of the total age of the two heads of the
household or the age of the single head of household.
FHH Based on Married and F_Rel, it assumes the value 1 if there is only a female head
of household, 0 otherwise.
3
Wrangled data are available upon request.

APPENDIX C: R CODE USED IN THE PROJECT

We first import the ERIMData into a data frame and label it myData. The
fol lowing is a portion of the R code used for data wrangling and selected
business analytic models.1

Data Wrangling
myData$Yogurt <- ifelse(myData$YogExp > 0, 1, 0)

1
The master code for a comprehensive list of analytical techniques, including the ones not detailed in
the article, is available upon request.
632 Teaching Business Analytics

myData$Dinner <- ifelse(myData$DinExp > 0, 1, 0)

myData$Sglfam <- ifelse(myData$ResType = = 3, 1, 0)
myData$Ownhome <- ifelse(myData$ResStatus = = 1, 1, 0) myData$Pets <-
ifelse((myData$Cats + myData$Dogs) > 0, 1, 0) myData$Married <-
ifelse(myData$M_Rel > 0 & myData$F_Rel > 0, 1, 0) myData$EduBoth <-
ifelse(myData$MEdu > = 9 & myData$FEdu > = 9, 1, 0) myData$EduOne <-
ifelse(myData$Married = = 0, ifelse(myData$MEdu> = 9 | myData$FEdu> =
9, 1,0),0)
myData$WrkHrs<- ifelse(myData$Married = = 1,
0.5*(myData$MWrkHrs+my Data$FWrkHrs),
(myData$MWrkHrs+myData$FWrkHrs))
myData$HHMembers <- ifelse(myData$Married = = 1, myData$HHNbr
2,myData$HHNbr-1)
myData$Age <- ifelse(myData$Married = = 1, (1985 -
0.5*(myData$MBirth+myData$FBirth)),(1985-
(myData$MBirth+my Data$FBirth)))
myData$FHH<- ifelse(myData$Married = = 1, 0, myData$F_Rel) myData =
subset(myData, select = -c(ResType, ResStatus, HHNbr, MWrkHrs, MEdu,
FWrkHrs, FEdu, FBirth, F_Rel, MBirth, M_Rel, Cats, Dogs))
summary(myData$YogExp)
summary(myData$DinExp)
YogYes <- myData[myData$Yogurt = = 1,]
YogNo <- myData[myData$Yogurt = = 0,]
DinYes <- myData[myData$Dinner = = 1,]
DinNo <- myData[myData$Dinner = = 0,]
data.frame(colMeans(myData), colMeans(YogYes), colMeans(YogNo),
colMeans(DinYes), colMeans(DinNo))
data.frame(nrow(myData), nrow(YogYes), nrow(YogNo), nrow(DinYes),
nrow(DinNo))

Classification Model (Logistic Regression) for Frozen Dinners

Packages ‘caret’, ‘gains’, ‘pRoc’, must be installed and loaded before running
the code.
myData$BinDin <- as.factor(myData$Dinner)
myDataD <- myData[, -c(1, 4, 5, 6, 7)]
set.seed(1)
myIndex <- createDataPartition(myDataD$BinDin, p = 0.6, list =
FALSE) trainSet <- myDataD[myIndex,]
validationSet <- myDataD[-myIndex,]
Logit_Reg <- glm(BinDin ∼ ., data = trainSet, family = “binomial”)
Logit_Reg_Pred <- predict(Logit_Reg, newdata = validationSet, type = “re
sponse”)
confusionMatrix(as.factor(ifelse(Logit_Reg_Pred>0.5,1,0)), validationSet$Bin
Din, positive = ‘1’)
confusionMatrix(as.factor(ifelse(Logit_Reg_Pred>0.33,1,0)),
validationSet$Bin Din, positive = ‘1’)
validation_BinDin <- as.numeric(as.character(validationSet$BinDin))
Jaggia, Kelly, Lertwachara, and Chen 633

gain_table <- gains(validation_BinDin, Logit_Reg_Pred)

gain_table
plot(c(0, gain_table$cume.pct.of.total*sum(validation_BinDin)) ∼ c(0,
gain_table$cume.obs), xlab = “# cases”, ylab = “Cumulative”, main = “Lift
Chart”, type = “l”)
lines(c(0, sum(validation_BinDin)) ∼ c(0, dim(validationSet)[1]), lty =
2) heights <- gain_table$mean.resp/mean(validation_BinDin)
dwlc <- barplot(gain_table$mean.resp/mean(validation_BinDin), names.arg =
gain_table$depth, ylim = c(0,2), xlab = “Percentile”, ylab = “Mean
Response”, main = “Decile-Wise Lift Chart”)
r<- roc(validation_BinDin, Logit_Reg_Pred)
plot.roc(r)
auc(r)
 Prediction Model (Regression Tree) for Frozen Dinners Packages
‘rpart’, ‘rpart.plot’, and ‘gains’ must be installed and loaded before run ning
the code.
myData$LnDin <- log(1+myData$DinExp)
myDataP<- myData[, -c(1, 4, 5, 6, 7, 18)]
set.seed(1)
myIndex <- createDataPartition(myDataP$LnDin, p = 0.6, list =
FALSE) trainSet <- myDataP[myIndex,]
validationSet <- myDataP[-myIndex,]
default_tree <- rpart(LnDin ∼ ., data = trainSet, method =
“anova”) summary(default_tree)
prp(default_tree, type = 1, extra = 1, under = TRUE, varlen =
10) predicted_value_valid <- predict(default_tree,
validationSet)
accuracy(predicted_value_valid, validationSet$LnDin)
gain_table <- gains(validationSet$LnDin,
predicted_value_valid) gain_table
plot(c(0, gain_table$cume.pct.of.total*sum(validationSet$LnDin)) ∼ c(0,
gain_table$cume.obs), xlab = “# cases”, ylab = “Cumulative”, main = “Lift
Chart”, type = “l”)
lines(c(0, sum(validationSet$LnDin)) ∼ c(0, dim(validationSet)[1]), lty = 2)
heights <- gain_table$mean.resp/mean(validationSet$LnDin) dwlc<-
barplot(gain_table$mean.resp/mean(validationSet$LnDin), names.arg =
gain_table$depth, ylim = c(0,2.5), xlab = “Percentile”, ylab = “Mean Re
sponse”, main = “Decile-Wise Lift Chart”)

Sanjiv Jaggia is a Professor of economics and finance at California

Polytechnic State University. Dr. Jaggia earned a PhD in economics from
Indiana University and is a Chartered Financial Analyst. His scholarly
activities include research in applied econometrics and finance and successful
textbooks in business statistics and business analytics.
634 Teaching Business Analytics

Alison Kelly is a Professor of economics at Suffolk University. Dr. Kelly

earned a PhD in economics from Boston College and is a Chartered Financial
Analyst. Her scholarly work includes research in economics and successful
textbooks in business statistics and business analytics.

Kaveepan (Kevin) Lertwachara is a Professor of Information Systems at

Califor nia Polytechnic State University. Dr. Lertwachara earned a PhD in
Operations and Information Management from the University of Connecticut.
His scholarly work includes research on analytics and electronic commerce
and a successful textbook in business analytics.

Leida Chen is a Professor of Information Systems at California Polytechnic

State University. Dr. Chen earned a PhD in Management Information Systems
from Uni versity of Memphis. He has published in leading academic journals
and coauthored a book on mobile application development and a textbook in
business analytics.