Fig. 4. This ﬁgure depicts a potential scenario for bacterial
classiﬁcation. The pond sample is placed under a microscope
with a video sensor attached to the lens. This sensor is con-
nected to a large monitor that displays the magniﬁed pond
sample. The data extraction from each object could be com-
pleted by the students or hypothetically by image processing
software. Either way this data is input to the k-means al-
gorithm. The output of the algorithm will specify to which
group each bacterium belongs.
features be used? What about using more than three features?
Can we visualize this data? Why or why not?
Each approach undoubtedly has its own strengths and
weaknesses to be highlighted. In the recycling example the
containers are all analyzed before each is deposited in its
appropriate bin. How does this fair on the design of the
conveyor belt? In the bacterial classiﬁcation example the stu-
dents had to deal with a bacterium that has both animal and
plant features. How did this bacterium group? Was it consis-
tent? Could they infer whether it is more of one class over the
other? What about automating the feature extraction? Could
they develop a ML technique to automatically measure the
shape, size and motility of each bacterium?
Another important concept is that of generality. The stu-
dents can run their ML design with different datasets, but
remain using the same parameters and features. Based on
the evaluation criteria does each run perform similarity or
are some much better than others? What impact would this
have if they were to spend a large sum of money on an unpre-
6. STRUCTURE OF LAB IMPLEMENTATION
We propose a variety of methods for presenting the lab to the
students, although any combination of pedagogy may be cho-
sen for a particular classroom. Our past experience designing
and implementing signal processing labs on topics such as im-
age processing and bioinformatics have taught us that the stu-
dents tend to respond best to these topics when they are ﬁrst
briefed on the lab and background followed by a short, open
class discussion . In this ML lab we suggest class par-
ticipation when walking through the example exercise such
as the coin-sorting problem. The students may then break
off into small groups to work on another exercise such as the
bacterial classiﬁcation problem. The results of each group’s
algorithm performance may then be compared and discussed
collectively as a class. The lab could then conclude with one
or more of the ideas from the suggested class discussion top-
7. FUTURE WORK
The scalable nature of the proposed ML lab suggests that ad-
ditional lab modules may be developed that expand on the
basic concepts described here. We envision labs utilizing neu-
ral networks to highlight advanced ML techniques as well as
provide insight into biologically inspired algorithms. We also
intend to develop an activity for use in classrooms where stu-
dents have chosen to focus on ﬁelds stemming from the cre-
ative arts. We advocate student exposure to these topics dur-
ing secondary education because not only is this lab activity
an introduction to engineering, it is insight into how many de-
cisions are made on a daily basis across virtually all areas of
 Richard Duda, Peter Hart, and David Stork, Pattern
Classiﬁcation, Wiley, New York, 2nd edition, 2001.
 Christopher M. Bishop, Pattern recognition and ma-
chine learning, Springer, New York, 1st edition, 2006.
 JJ Hopﬁeld, “Neural networks and physical systems
with emergent collective computational abilities,” Pro-
ceedings of the National Academy of, vol. 79, no. April,
pp. 2554–2558, 1982.
 Patrick K. Simpson, Ed., Neural Network Applications,
IEEE Press, Piscataway, NJ, 1st edition, 1997.
 P Chan and S Stolfo, “Toward scalable learning with
non-uniform class and cost distributions: A case study
in credit card fraud detection,” Proc. of the Fifteenth Na-
tional Conference on Artiﬁcial Intelligence, vol. AAAI-
98, pp. 164–168, 1998.
 Kou-yuan Huang, Senior Member, and Wen-lung
Chang, “A Neural Network Method for Prediction of
2006 World Cup Football Game,” Proc. of International
Joint Council on Neural Networks, pp. 259–266, 2010.
 M Egmont-Petersen, D. de Ridder, and H. Handles, “Im-
age processing with neural networks—a review,” Pat-
tern Recognition, vol. 35, no. 10, pp. 2279–2301, Oct.