Unsupervised Learning

Wei Wu
May 11, 2022

1
Contents
1 Overview of machine learning 3

2 What is Unsupervised Learning? 3

2.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.3 Di↵erences compare with supervised learning . . . . . . . . . . 5
2.4 Advantages . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

3 Types Of Algorithms(commonly used) 7

3.1 Clustering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.1.1 K-means Clustering . . . . . . . . . . . . . . . . . . . . 7
3.1.2 Hierarchical Clustering . . . . . . . . . . . . . . . . . . 8
3.1.3 K nearest neighbors . . . . . . . . . . . . . . . . . . . . 10
3.2 Association Analysis . . . . . . . . . . . . . . . . . . . . . . . 11
3.2.1 Association Rules . . . . . . . . . . . . . . . . . . . . . 11
3.2.2 Apriori Algorithm . . . . . . . . . . . . . . . . . . . . . 12

4 Applications 17

5 Summary 18

6 Sources 19

2
1 Overview of machine learning
Supervised learning: Input is provided as a labelled dataset, a model can
learn from it to provide the result of the problem easily.
Unsupervised learning: There is no complete and clean labelled dataset.
Reinforcement learning: The algorithms learn to react to an environment on
their own.
Semi- supervised learning: It is an approach to machine learning that com-
bines a small amount of labeled data with a large amount of unlabeled data
during training.

2 What is Unsupervised Learning?

In supervised learning, we aim to train a model to be capable of mapping
an input to output after learning some features, acquiring a generalization
ability to correctly classify never-seen samples of data. But sometimes we
don’t know what is the output, as we only have the input data and we can’t
define an output label for each input sample. Let’s suppose we’re working
for a company that sells clothes and we have data from previous customers:
how much they spent, their ages and the day that they bought the product.
Our task is to find a pattern or relationship between the variables in order to
provide the company with useful information so they can create marketing
strategies, decide on which type of client they should focus on to maximize
the profits or which customer segment they can put more e↵ort to expand in
the market.

3
In this case, the output will not be a label since we can’t model our needs
in this way. Instead, our program should be able to group the customers
accordingly to what makes them similar or unique. This grouping will be
conducted from the features learned during the training phase, and in this
case, we have unsupervised learning approach since there is no supervisor to
provide labels for the inputs to map them to the output.

2.1 Definition
Unsupervised machine learning is the process of inferring underlying hidden
patterns from historical data. Within such an approach, a machine learning
model tries to find any similarities, di↵erences, patterns, and structure in
data by itself. No prior human intervention is needed.

2.2 Examples
1.Picture a toddler. The child knows what the family cat looks like (provided
they have one) but has no idea that there are a lot of other cats in the world
that are all di↵erent. The thing is, if the kid sees another cat, he or she will
still be able to recognize it as a cat through a set of features such as two ears,
four legs, a tail, fur, whiskers, etc.

4
2.We have the following items and want to cluster them with unsupervised
learning. They are clustered by some di↵erent characteristics.

3.We have the following animals and try to cluster them with the help of
unsupervised learning.

2.3 Di↵erences compare with supervised learning

The key di↵erence is that with supervised learning, a model learns to pre-
dict outputs based on the labeled dataset, meaning it already contains the
examples of correct answers carefully mapped out by human supervisors.
Unsupervised learning, on the other hand, implies that a model swims in
the ocean of unlabeled input data, trying to make sense of it without human
supervision.

5
2.4 Advantages
Why are we using Unsupervised Learning?

• Unsupervised learning is helpful for data science teams that don’t know
what they’re looking for in data. It can be used to search for unknown
similarities and di↵erences in data and create corresponding groups.
For example, user categorization by their social media activity.

• The given method doesn’t require training data to be labeled, saving

time spent on manual classification tasks.

• Unlabeled data is much easier and faster to get.

• It reduces the chance of human error and bias, which could occur during
manual labeling processes.

6
3 Types Of Algorithms(commonly used)
3.1 Clustering
Clustering automatically categorizes data into groups according to similarity
criteria.
For example, We need to arrange the following blocks of shapes and colors.
According to di↵erent characteristics of this blocks, we could have two dif-
ferent results.

3.1.1 K-means Clustering

An iterative algorithm that categorizes data into a predefined number of
groups or clusters.
(1) (1)
Given an inital set of k means m1 ,...,mk ,the algorithm proceeds by alter-
nating between two steps:
Assignment Step: It measures distances between k centroids and individual
data points.
(t) (t) (t)
Si = {xp : ||xp mi ||2  ||xp mj ||2 8j, 1  j  k},
and each xp is assigned to exactly one S (t) .
Update Step: Then, it updates the position of each centroid as the mean of
respective data P
points belonging to each cluster.
(t+1)
mi = 1(t) xj
|Si | (t)
xj 2Si

7
It repeats the process until no centroid moves more than a given threshold.

The algorithm has converged when the assignments no longer change.It is

often presented as assigning objects to the nearest cluster by distance.
Advantages: Guarantees convergence. Can warm-start the positions of cen-
troids. Generalizes to clusters of di↵erent shapes and sizes, such as elliptical
clusters.
Disadvantages: Firstly, you have to select how many groups/classes there are.
This isn’t always trivial and ideally with a clustering algorithm we’d want
it to figure those out for us because the point of it is to gain some insight
from the data. K-means also starts with a random choice of cluster centers
and therefore it may yield di↵erent clustering results on di↵erent runs of the
algorithm. Thus, the results may not be repeatable and lack consistency.

3.1.2 Hierarchical Clustering

An iterative algorithm that categorizes data into hierarchical groupings. It
has 2 categories: top-down(divisive) or bottom-up(agglomerate).
(1) Agglomerative Hierarchical Clustering: Bottom-up algorithms treat each
data point as a single cluster at the outset and then successively merge pairs
of clusters until all clusters have been merged into a single cluster that con-
tains all data points.

8
Example:

The example shows how six di↵erent clusters(data points) are merged step
by step on distance until they all create one large cluster.
How to compute the distance between clusters that contain more than one
data object? We have four methods below:
Centroid: Distance between the centroids of the two clusters.
Average Linkage: Average Distance between all pairs of points of the two
clusters.
Single Linkage: Distance between the two most similar data objects of the
two clusters.
Complete Linkage: Distance between the two most dissimilar data objects of
the two clusters.

(2) Divisive Hierarchical Clustering: starts with the whole data set as a
single cluster and then divides clusters step by step into small clusters.

9
The advantage of Hierarchical Clustering is we don’t have to pre-specify the
clusters. However, it doesn’t work very well on vast amounts of data or huge
datasets. And there are some disadvantages of the Hierarchical Clustering
algorithm that it is not suitable for large datasets.

3.1.3 K nearest neighbors

It is the simplest of all machine learning classifiers. It works very well when
there is a distance between examples. The learning speed is slow when the
training set is large, and the distance calculation is nontrivial.
Example:

The test sample (green dot) should be classified either to blue squares or
to red triangles.
If k = 3 (solid line circle) it is assigned to the red triangles because there are
2 triangles and only 1 square inside the inner circle.
If k = 5 (dashed line circle) it is assigned to the blue squares (3 squares vs.
2 triangles inside the outer circle).

Advantages of KNN:
1. No Training Period: KNN is called Lazy Learner. There is no training
period for it. It stores the training dataset and learns from it only at the time

10
of making real time predictions. So it is much faster than other algorithms.
2.Since the KNN algorithm requires no training before making predictions,
new data can be added seamlessly which will not impact the accuracy of the
algorithm.
Disadvantages of KNN:
1. Does not work well with large dataset: In large datasets, the cost of cal-
culating the distance between the new point and each existing points is huge
which degrades the performance of the algorithm.
2. Does not work well with high dimensions: The KNN algorithm doesn’t
work well with high dimensional data because with large number of dimen-
sions, it becomes difficult for the algorithm to calculate the distance in each
dimension.

3.2 Association Analysis

Association Analysis is the task of finding interesting relationships in large
datasets. And there are two forms to look for the relationship: Association
rules and Frequent Itemset .

3.2.1 Association Rules

Association rules allow you to establish associations among data objects in-
side large databases. It is descriptive not the predictive method, generally
used to discover the relationships between variables in large databases. For
example, people that buy a new home most likely to buy new furniture. In
addition, Amazon will recommend you some sport clothes if the last search
was related to sports.
The strength of a given association rule is measured by two main parameters:
support and confidence. Support refers to how often a given rule appears in
the database being mined. Confidence refers to the amount of times a given
rule turns out to be true in practice.

Support and Confidence

Definition 1: Support of itemset A, denoted support(A)= frequency an Item-
set A appears in the dataset T:
support(A)= |{t2T|T:A✓t}|
|
, t is an itemset.
Definition 2: Let A and B be two itemsets. An association rule A ! B

11
asserts that if a transaction contains A, it is also likely to contain B.
Definition 3: The support of an association rule A ! B is |AB|.
Definition 4: The Confidence is the percentage of all transactions satisfying
X that also satisfy Y.
Definition 5: The confidence of an association rule A ! B is |AB| |A|
:
supp(A\B) number of transactions containing A and B
conf(A ) B) = supp(A)
= number of transactions containing A

Example:
We have Itemsets: A={apple}, B={peach and apple}, C={peach and ba-
nana}, D={strawberry}, E={banana}
Dataset T={A,B,C,D,E}
Then we have the following information:
support(A)= 25
conf(A ) B)= 12

Advantages of association rules : There are many benefits of using As-

sociation rules like finding the pattern that helps understand the correlations
and co-occurrences between data sets.

Disadvantages of association rules :

(1) It requires high computation if the itemsets are very large and the mini-
mum support is kept very low.
(2) The entire database needs to be scanned.

3.2.2 Apriori Algorithm

Apriori Algorithm uses frequent itemsets to generate association rules and
it is based on the Apriori principle mentioned above. Apriori algorithms are
designed to process databases containing transactional information content
(e.g., lists of items purchased by customers).

Frequent itemset
Frequent itemset is an itemset whose support value is greater than threshold
value . Which means: support(X) minsup.

Example on finding Frequent itemsets:

12
Consider the given dataset with given transactions.

We can try to calculate the support value of di↵erent itemsets.

support(A)= 35 , support(D)= 15 , ...
Let minsup= 25 , then we have :

Total itemsets: 31 (= 25 - 1), 5 is the amount of single item, which means

A,B,C,D,E in this example.
Amount of Frequent itemsets: 15
Amount of Infrequent itemsets: 16

13
We can try to avoid considering all the items based on Apriori Principle:
(1) If an itemset is infrequent then all its supersets are infrequent.
(2) If an itemset is frequent then all its subsets are frequent.
(What is superset? - The opposite of subset
A is a superset of another set B if all elements of the set B are elements of
the set A. The superset relationship is denoted as A B. Opposing we say,
B is the subset of A.)
With the help of these two principles, we can try to make the algorithm
easier, for example:
Itemset D is infrequent, so all its supersets are infrequent. (Which means
all the itemsets with element D are infrequent, so we can ignore it and keep
looking for frequent itemsets.)
Besides frequent itemsets, two more important itemsets will be introduced
below:

Closed itemset
An itemset X is frequent and no immediate superset of X has same support
as X.

Maximal itemset
An itemset X is frequent and no immediate supersets of X are frequent.

14
In this example we can find out that for the frequent itemset C with supp(C)= 45 ,
none of its supersets have same support as C, so itemset C is closed.
And the frequent itemset {A, B, C, E} is the only maximal itemset because
it doesn’t have (frequent) supersets.
In conclusion, F=15, C=5, M=1 .
So the relationship between these three itemsets is: M ⇢ C ⇢ F .

So how does Apriori Algorithm actually work? - It will be shown in the

following example:
We have a dataset T with 4 transactions, let minsup = 24 = 12 , mincof = 12 .

Step 1: Scan T, calculate the support for each candidates, then we can get
C1 (the candidate set of level k) and L1 (the frequent dataset).

C1 L1
Step 2: Combine the itemsets in L1 to generate C2 .

15
Step 3: Scan C2 in order to get L2 .

C2 L2
Step 4: Combine itemsets in L2 to get C3 :

Step 5: Scan C3 .

C3 L3

) STOP, the algorithm can’t run anymore because no new frequent itemsets
are identified.
) {B, C, E} is the final frequent itemset.
) Calculate the confidence and compare with the mincof to find out which
association rules are possible.

When {B, C, E} is the frequent itemset. We have the following possibili-

ties for the association rules:
(1) B ^ C ! E , E ! B ^ C
(2) B ^ E ! C , C ! B ^ E
(3) C ^ E ! B , B ! C ^ E
Then we need to find out which possibilities are acceptable.

16
(1) conf(B ^ C ! E)= 22 = 1 , conf(E ! B ^ C) = 23
(2) conf(B ^ E ! C) = 23 , conf(C ! B ^ E) = 23
(3) conf(C ^ E ! B) = 22 = 1 , conf(is B ! C ^ E) = 23
All confidences are higher than minconf, so all of the possibilities are accept-
able.

Advantages of Ariori Algorithm : Easy to implement. Use large

itemset property.

Disadvantages of Apriori Algorithm : Requires many database scans.

Very slow.

4 Applications
Applications of Apriori Algorithm :
1.In Education Field: Extracting association rules in data mining of admit-
ted students through characteristics and specialties.
2.In the Medical field: Analyzing the patient’s database.
3.In Forestry: Analyzing the probability and intensity of forest fire with the
forest fire data.
4.Apriori is used by many companies like Amazon in the Recommender Sys-
tem and by Google for the auto-complete feature.

Application of association rules :

Medicine uses Association rules to help diagnose patients. When diagnosing
patients there are many variables to consider as many diseases will share sim-
ilar symptoms. With the use of the Association rules, doctors can determine
the conditional probability of an illness by comparing symptom relationships
from past cases.

17
5 Summary
• Unsupervised learning is a machine learning technique, where you do not
need to supervise the model.
• Unsupervised machine learning helps you to finds all kind of unknown pat-
terns in data.
• Clustering and Association are two types of Unsupervised learning.
• Three types of clustering methods are 1) k-means 2) Agglomerative 3) k
nearest neighbors.
• Association rules allow you to establish associations amongst data objects
inside large databases.
• In Supervised learning, Algorithms are trained using labelled data while
in Unsupervised learning Algorithms are used against data which is not la-
belled.
• The biggest drawback of Unsupervised learning is that you cannot get pre-
cise information regarding data sorting.

18
6 Sources
1.https://www.altexsoft.com/blog/unsupervised-machine-learning/

2.https://en.wikipedia.org/wiki/K-meansclustering

3.https://en.wikipedia.org/wiki/Hierarchicalclustering

4.https://www.youtube.com/watch?v=FvFsjdMwANU

5.https://www.guru99.com/unsupervised-machine-learning.html

6.https://medium.com/@manilwagle/association-rules-unsupervised-
learning-in-retail-69791aef99a

7.https://datascienceguide.github.io/association-rule-mining

8.https://en.wikipedia.org/wiki/Associationrulelearning

9.https://www.softwaretestinghelp.com/apriori-algorithm/

10.https://www.youtube.com/watch?v=nSpajfE5Ujc

11.https://bainingchao.github.io/2018/09/27/Apriori/

12.https://www.youtube.com/watch?v=rZRxCpLdNrg

13.https://www.aitude.com/supervised-vs-unsupervised-vs-reinforcement/

14.http://theprofessionalspoint.blogspot.com/2019/02/advantages-
and-disadvantages-of-knn.html

19
15.https://www.google.com.hk/search?q=kclient=safarisxsrf=APq-
WBsErrH
HV9eY41bPBOnp43bCVA51Mw:1649873895399source=lnmstbm=ischsa=Xved
=2ahUKEwiA89Wb05H3AhUMgP0HHbYnAuoQAUoAXoECAIQAwbiw=883bih
=708dpr=2imgrc=wz72rE1JY6tXM

16.https://www.iguazio.com/glossary/unsupervised-ml/?utmsource=adword
sutmterm=utmcampaign=GGl-DSA-ROWutmmedium=ppcutmcontent=
MachineLearninghsacam=11602556551hsaver=3hsamt=hsasrc=ghsagrp=
108334939330hsakw=hsanet=adwordshsaacc=3049108309hsaad=479289328973
hsatgt=dsa-392284169515gclid=CjwKCAjwo8-SBhAlEiwAopc9W6okrua9onhT8F
zBIuHRY6mxp7ASfxXKujZA9Z26JmQysAPzfSDOGBoCncAQAvDBwE

17.https://www.baeldung.com/cs/examples-supervised-unsupervised-
learning

20