Machine Learning Techniques
Machine Learning Techniques
net/publication/353887208
CITATIONS READS
2 751
3 authors:
SEE PROFILE
All content following this page was uploaded by Asongo Abraham Iorkaa on 13 August 2021.
ABSTRACT: Major companies have actively reoriented themselves around Artificial Intelligence and
machine learning. A lot of resources are being deployed and attention focused on the use of machine learning in
a bid to convincing the world that the machine intelligence revolution is arriving now. The technological
transformation resulting to powering new self-driving cars, virtual assistants, disease detection and therapy
planning and many more are just few out of numerous applications of machine learning. This paper, through a
review of the available literature seeks to offer conceptual and practical insights on the techniques, methods and
algorithms of machine Learning. This paper has shown that there are numerous research on applications of
machine learning in the management of banking risks such as credit risk, market risk, operational risk and
liquidity risk and other fields has been explored; however, it doesn’t appear commensurate with the needed
knowledge on the general models, application software and the applications in other sectors. This paper
therefore shed more light that could significantly add to the knowledge of the field by giving insight on the
history, types, models and other areas of applications of machine Learning.
Keywords: Algorithms, Applications, Machine Learning, Techniques, Training,
---------------------------------------------------------------------------------------------------------------------------------------
Date of Submission: 24-07-2021 Date of Acceptance: 09-08-2021
---------------------------------------------------------------------------------------------------------------------------------------
training algorithms such that they can learn how to networks; these were mostly perceptron and other
make decisions. Training in machine learning entails models that were later found to be reinventions of the
giving a lot of data to the algorithm and allowing it to generalized linear models of statistics. Probabilistic
learn more about the processed information. reasoning was also employed, especially in
To explain further how a machine learns automated medical diagnosis.
from data, let us look at this example. Supposing we Machine learning, reorganized as a separate
have a table comprising of different fruits to be field, started to flourish in the 1990s. The field
differentiated based on their weight and colour with changed its goal from achieving artificial intelligence
the last row showing only the weight and colour as to tackling solvable problems of a practical nature. It
shown in table 1 below: shifted focus away from the symbolic approaches it
had inherited from AI, and toward methods and
Table 1: Machine Learning algorithm explained models borrowed from statistics and probability
Weight Colour Type of theory[7].
(grams) Fruit Machine learning and data mining often
80 Green Apple employ the same methods and overlap significantly.
85 Green Apple The two areas overlap in many ways: data mining
95 Green Apple uses many machine learning methods, but often with
90 Yellow Orange a slightly different goal in mind. Machine learning
100 Yellow Orange also has intimate ties to optimization: many learning
103 Yellow ?? problems are formulated as minimization of some
loss function on a training set of examples. Loss
Here a machine learning algorithm can be functions express the discrepancy between the
developed to try to identify whether the fruit is an predictions of the model being trained and the actual
orange or an apple. After the algorithm is fed with the problem instances (for example, in classification, one
training data, it will learn the differing characteristics wants to assign a label to instances, and models are
between an orange and an apple. Therefore, if trained to correctly predict the pre-assigned labels of
provided with data of weight and colour, it can a set examples). The difference between the two
predict accurately the type of fruit with those fields arises from the goal of generalization: while
characteristics. optimization algorithms can minimize the loss on a
training set, machine learning is concerned with
minimizing the loss on unseen samples [8]
II. THE HISTORY OF MACHINE
LEARNING III. TYPES OF MACHINE LEARNING
The word Machine learning was first coined
[1] categorized machine learning algorithms
by Arthur Samuel in 1952 [5]. In 1957, Frank into supervised, unsupervised and reinforcement
Rosenblatt – at the Cornell Aeronautical Laboratory –
learning algorithms: Figure one present the
combined Donald Hebb's model of brain cell
classification in a pictorial form:
interaction with Arthur Samuel's Machine Learning
efforts and created the perceptron. In 1967, the
nearest neighbor algorithm was conceived, which
was the beginning of basic pattern recognition. This
algorithm was used for mapping routes and was one
of the earliest algorithms used in finding a solution to
the traveling salesperson’s problem of finding the
most efficient route. In the 1960s, the discovery and
use of multilayers opened a new path in neural
network research. It was discovered that providing
and using two or more layers in the perceptron
offered significantly more processing power than a
perceptron using one layer [5].
According to [6]machine learning grew out
Figure 1: Types of Machine Learning
of the quest for artificial intelligence. Already in the
early days of artificial intelligence as an academic
i. Supervised Learning
discipline, some researchers were interested in having
Supervised learning is a core area of
machines learn from data. They attempted to machine learning. In supervised learning the goal is
approach the problem with various symbolic
to learn a mapping from the input to the output. The
methods, as well as what were then termed neural
input is data that describes a collection of individual
objects of interest and are commonly referred to as association rule learning problem is where you want
instances or examples. The output is some outcome to discover rules that describe large portions of your
or result provided by a supervisor. Classification is a data, such as people that buy X also tend to buy Y.
form of supervised learning whereby a mapping (or Some popular examples of unsupervised learning
discriminant function) separates diff erent classes of algorithms are k-means for clustering problems and
the instances. The diff erent classes are specified by Apriori algorithm for association rule learning
the output which, in machine learning, is termed as problems.
the class label. The discriminant function is referred
to as a classifier or a model. A set of instances with iii. Reinforcement machine learning algorithms
their known class label is termed a training set. Reinforcement machine learning algorithms is a
During classification, a model is defined by a set of learning method that interacts with its environment
parameters that are optimized to generate a mapping by producing actions and discovers errors or rewards.
from training set instances to training set labels. The Trial and error search and delayed reward are the
trained model can be used to classify or label new, most relevant characteristics of reinforcement
unseen instances. learning. This method allows machines and software
The majority of practical machine learning agents to automatically determine the ideal behavior
uses supervised learning. Supervised learning is within a specific context in order to maximize its
where you have input variables (x) and an output performance. Simple reward feedback is required for
variable (Y) and you use an algorithm to learn the the agent to learn which action is best; this is known
mapping function from the input to the output. as the reinforcement signal. When it comes to
Y = f(X) (1) explaining machine learning to those not concerned
The goal is to approximate the mapping function so in the field, reinforcement learning is probably the
well that when you have new input data (x) that you easiest sub-field for this challenge. To make it more
can predict the output variables (Y) for that data. It is practical, it can be said Reinforcement Learning is
called supervised learning because the process of an like teaching your dog (or cat if you live your life in a
algorithm learning from the training dataset can be challenging way) to do tricks: you provide goodies as
thought of as a teacher supervising the learning a reward if your pet performs the trick you desire,
process. We know the correct answers, the algorithm otherwise, you punish him by not treating him, or by
iteratively makes predictions on the training data and providing lemons. Dogs really hate lemons. Beyond
is corrected by the teacher. Learning stops when the controversy, Reinforced learning is a more complex
algorithm achieves an acceptable level of and challenging method to be realized, but basically,
performance. it deals with learning via interaction and feedback, or
in other words learning to solve a task by trial and
ii. Unsupervised Learning error, or in other-other words acting in an
According to [9] this machine learning environment and receiving rewards for it. Essentially
algorithms are used when the information used to an agent (or several) is built that can perceive and
train is neither classified nor labeled. Unsupervised interpret the environment in which is placed,
learning studies how systems can infer a function to furthermore, it can take actions and as well as interact
describe a hidden structure from unlabeled data. The with the environment.
system doesn’t figure out the right output, but it
explores the data and can draw inferences from IV. MACHINE LEARNING MODELS
datasets to describe hidden structures from unlabeled This session discussed various machine Learning
data. Models. These includes:
Unsupervised learning is where you only
have input data (X) and no corresponding output i. Decision tree methodology
variables. The goal for unsupervised learning is to Decision trees is one of the machine
model the underlying structure or distribution in the learning models. According to [10], decision tree
data in order to learn more about the data. These are approach or the recursive partitioning algorithm
called unsupervised learning because unlike (RPA) is a non- parametric, complex and
supervised learning above there is no correct answers computerized intensive sorting algorithm. The basic
and there is no teacher. Algorithms are left to their idea is to split the sample responses into the new sub-
own devises to discover and present the interesting samples that are as homogeneous as possible and as
structure in the data.Unsupervised learning problems different from each other, and then to repeatedly split
can be further grouped into clustering and association the sub-sample into subgroups until it generates the
problems. A clustering problem is where you want to possibility for decision- making. The entire sample is
discover the inherent groupings in the data, such as the root node, while the sub-samples are called
grouping customers by purchasing behavior while an nodes. Figure 2 is a good example of decision tree
showing credit risk classification into bad credit and absolute errors or minimizing the maximum error
good credit. [12]. To solve the problem by linear programming, a
popular formulation is asfollows:
(4)
≥
when such activities are reported and counted to be operate a device, perform commands, or write
true, they help to improve the surveillance services. without having to use a keyboard, mouse, or press
This happens with machine learning doing its job at any buttons. All commercial purpose speech
the backend. A small video file contains more recognition system uses a machine learning approach
information compared to text documents and other to recognize the speech. This is because the speech
media files such as audio, images. For this reason, recognition system using machine learning approach
extracting useful information from video, i.e., the outperforms better than the speech recognition
automated video surveillance system has become a system using a traditional method.
hot research issue. With this regard, video
surveillance is one of the advanced application of a vi.Medical Services
machine learning approach. A system with the ability Machine learning methods, tools are used
to gather information about the presence of the same extensively in the area of the medical-related
person in a different frame of a video is highly problem. As an instance to detect a disease, therapy
demanding. There are several methods of machine planning, medical-related research, prediction of the
learning algorithm to track the movement of human disease situation.The value of machine learning in
and identifying them. healthcare is its ability to process huge datasets
beyond the scope of human capability, and then
iii. Email Spam and Malware Filtering reliably convert analysis of that data into clinical
There are a number of spam filtering insights that aid physicians in planning and providing
approaches that email clients use. To ascertain that care, ultimately leading to better outcomes, lower
these spam filters are continuously updated, they are costs of care, and increased. Some of the notable
powered by machine learning. When rule-based spam areas of applications in medical services include
filtering is done, it fails to track the latest tricks Identifying diseases and diagnosis, drug discovery
adopted by spammers. and manufacturing, medical imaging diagnosis,
To classify email and filter the spam in an personalized medicine, machine learning-based
automatic way machine learning algorithm is behavioral modification, smart health records,
employed. There are many techniques such as multi- clinical trial and research and crowdsourced data
layer perception, C4.5 decision tree induction etc. collection.
used to filter the spam. The rule-based spam filtering
has some drawbacks to filter the spam, whereas spam vii. Online Customer Support
filtering using the machine Learning approach is Online Customer Services are automated
more efficient. services that helps customers solve problems. Such a
service is usually connected to a business or brand
iv. Online Fraud Detection which sells products and offers support. There are
Machine learning is proving its potential to different types of online customer service, depending
make cyberspace a secure place and tracking on the type of industry involved. Recently almost all
monetary frauds online is one of its good websites allow the customer to chat with the website
applications. Online fraud detection is an advanced representative. However, not website has an
application of machine learning algorithm. This executive. Basically, they develop a chat-bot to chat
approach is practical to provide cybersecurity to the with the customer to know their opinion. This is
users efficiently. Recently, PayPal is using a machine possible only for the machine learning approach. It’s
learning and artificial intelligence algorithm for just a beauty of machine learning algorithm.
money laundering. This advanced machine learning
and artificial intelligence example helps to reduce the viii. Robot Control
loss and maximize the profit. Using machine learning Robotic control is the system that
in this application, the detection system becomes contributes to the movement of robots. This involves
robust than any other traditional rule-based system. the mechanical aspects and program systems that
makes possible to control robots. A machine learning
v. Speech Recognition algorithm is used in a variety of robot control system.
One of the applications of Machine Learning For instance, recently, several types of research have
is speech recognition. Speech recognition is the been working to gain control over stable helicopter
ability of a machine or program to identify words and flight and helicopter aerobatics. Another good
phrases in spoken language and convert them to a example of robotic control is the Darpa-sponsored
machine-readable format. Alternatively referred to as, competition- a robot driving for over one hundred
voice recognition is a computer software program or miles within the desert was won by a robot that used
hardware device with the ability to decode the human machine learning to refine its ability to notice distant
voice. Voice recognition is commonly used to objects (Mehedi, 2019).
credit risk. The Statistician, 45(1), 77– [30]. Battiti, Roberto; Mauro Brunato (2014). The
95. LION way. Machine Learning plus Intelligent
[14]. Feiring, B. R. (1986). Linear programming: Optimization.. Available online at
An Introduction. Beverly Hills: Sage https://www.academia.edu/28682987/ Viewed
Publications. on 29 May, 2021
[15]. Mangasarian O. (1965). Linear and nonlinear [31]. Neuro Networks (2020). Website information.
separation of patterns by linear programming, Accessed on April 3, 2021 from
Operations Research 13, pp. 444–452. www.neurosolutions.com/
[16]. Freed, N., and Glover, F. (1981). Simple but [32]. InfinityInsights (2021). Who we are: website
powerful goal programming models for Information accessed on July 6, 2021 from
discriminant problems. European Journal of www.infinityinsight.com/
Operational Research, 7, 44–60. [33]. Markus H., Ralf K. (2013). RapidMiner: Data
[17]. Ignizio, J. P. (1985). Introduction to Linear Mining Use Cases and Business Analytics
Goal Programming. Beverly Hills: Sage [34]. Applications (Chapman & Hall/CRC Data
Publications. Mining and Knowledge Discovery Series),”
[18]. Koehler, G., and Erenguc, S. (1990). CRC Press, October 25, 2013. Available on
Minimizing misclassifications in linear line at https://pdfs.semanticscholar.org/
discriminant analysis. Decision Downloaded on May 28, 2020
Sciences,21(1), 63–85. [35]. Delta Lake (2021). All your data, Analytics
[19]. Holland, J. (1975). Adaptation in Natural and and AI on one Lake Platform. website
Artificial Systems. Ann Arbor: University of information accessed from
Michigan Press. https://databricks.com/ on July 1, 2021.
[20]. Bishop C. M. (2006). Pattern Recognition and [36]. H2o Software. Open Source Machine Learning
Machine Learning. Springer. ISBN 0- 387- software. Website information accessed from
31073-8. www.h2o.ai/products/h2o/
[21]. Suykens, J. &Vandewalle, J. (1999). Least [37]. Mehedi H. (2020). Top 20 Best Examples and
squares support vector machines. Neural Applications of Artificial Intelligence and
processing letters, 9,293–300. 105 Machine Learning,Ubuntupit Blog.
[22]. Sprinthall, R. & Fisk, S. (1990). Basic https://www.ubuntupit.com/top-20-best-
statistical analysis. Prentice Hall, NJ, USA. 95 machine-learning-applications-in-real-world/
[23]. Baesens, B., Van Gestel, T., Stepanova, M., [38]. Hoy, Matthew B. (2018). "Alexa, Siri,
Suyken, J., and Vanthienen, J. (2003). Cortana, and More: An Introduction to Voice
Benchmarking state-of- the-art Assistants". Medical Reference
classification algorithms for credit scoring. Services Quarterly. 37 (1): 81–88.
Journal of the Operational Research Society, doi:10.1080/02763869.2018.1404391.
54(6),627– 635 PMID 29327988.
[24]. Breiman, L. (2001). Random forests. Machine
learning, 45, 5–32. 105
[25]. Breiman L. (2000). Randomizing outputs to
increase prediction accuracy. Machine
Learning, 40:229–242.
[26]. Witten, I., Frank. E., Hall M., Christopher J.
(2011). Data Mining: Practical machine
learning tools and techniques, 3rd Edition".
Morgan Kaufmann, San Francisco (CA).
Retrieved 2011-01-19
[27]. Salfold Systems (2020). Website information.
Accessed on June 10, 2021 from www.salford-
systems.com/
[28]. Adrian K. and Gary B. (2016). Learning
OpenCV 3: Computer Vision in C++ with the
OpenCV Library. O'Reilly Media.
pp. 26ff. ISBN 978-1-4919-3800-3.
[29]. Torch7 (2020): A Matlab-like Environment for
Machine Learning (PDF). Neural Information
Processing Systems. 2011. Online at
https://publications.idiap.ch/downloads/papers/