Algorithms for Machine Learning: Real-World Applications and

Future Research
Kundan Kumar(21MCA2029)

Abstract
The digital world has access to a multitude of data in the Fourth Industrial
Revolution (4IR, also known as Industry 4.0) era, including Internet of Things (IoT)
data, cybersecurity data, business data, mobile data, health data, social media
data, etc. Knowledge of artificial intelligence (AI), machine learning (ML), is
essential to intelligently assess these data and create the associated smart and
automated applications. Machine learning algorithms come in a variety of forms,
including supervised, unsupervised, semi-supervised, and reinforcement learning.
In addition, deep learning, a type of machine learning that encompasses a wider
range of techniques, can effectively examine a lot of data. We provide a thorough
overview of various machine learning techniques in this paper, showing how they
may be used to increase the functionality and intelligence of an application.
Determining the fundamentals of various machine learning approaches and how
they may be applied in a variety of real-world application areas, such as
cybersecurity systems, smart cities, healthcare, e-commerce, agriculture, and
many more, is thus the core contribution of this work. Based on our work, we also
identify the difficulties and promising paths for further research. Overall, this
article seeks to serve as a resource for decision-makers in a variety of real-world
settings and application areas, especially from a technical standpoint.

Keywords Machine learning: - Artificial intelligence, machine learning, deep

learning, data science, data-driven decision-making, predictive analytics, and
intelligent applications.

Introduction
According to Simon, learning is "the process of a change and enhancement in
behaviours through investigating new information through time." Machine
learning is the term used when the "learning" referred to in this definition is
carried out by computers. During the machine learning process, enhancement
refers to developing the optimum solution based on the previously acquired
knowledge and sample data (Srmaçek,2007). The term "big data" has emerged as
a result of advancements in information technologies. The concept of "big data"
is not new, and it can be characterised as vast, growing, unrestricted raw data
collections that cannot be analysed using conventional database procedures.
Large amounts of data are gathered through websites, ATMs, credit card swipers,
and other devices. The information gathered in this manner is ready for analysis.
Depending on the business industry, different data collection fields have distinct
objectives for analysis. Applications for machine learning can be found in a variety
of industries, including biology, calculative finance, automotive, aviation,
production, natural language processing, image processing, and computer vision.
The goal is predicated on the idea of analysing and interpreting the prior data,
though. Machine learning techniques and algorithms have been developed to
analyse and interpret data because humans are unable to do it (Amasyal,2008).
In this study, the recently popularised notion of machine learning is thoroughly
studied. The paper provides details on the development of machine learning, the
techniques and algorithms employed, and the domains in which it is applied. The
conclusion, which is the last section, includes the findings from the earlier
research.

Machine learning
Computer procedures that are based on an algorithm and follow specific steps
have no margin for mistake. In some instances, computers make judgments based
on the current sample data, which is different from commands that are created
to have a result depending on an input. In some circumstances, computers may
err in their decision-making just like people do. In other words, machine learning
is the process of giving computers the ability to learn using data and experience
much like a human brain (Gör,2014). The basic goal of machine learning is to
develop models that can learn from previous data to become better, recognise
intricate patterns, and find solutions to new issues.
Fig: -1 Machine Learning Types
• Supervised Learning: - It is a technique that generates the output set using
the current input data. Classification and regression supervised learning are
the two categories of supervised learning.
➢ Classification: - Dividing up the data into the groups listed on the
data set in accordance with their distinctive characteristics.
➢ Regression: - Estimating or drawing conclusions about the data's
additional characteristics based on the features it does have.
• Unsupervised Learning: - In unsupervised learning, there is no output data
provided, which is the difference between supervised and unsupervised
learning. The relationships and connections between the data are used in
the learning process. Unsupervised learning lacks training data as well.
➢ Clustering: - Finding the data groups that are like one another when
the data's intrinsic categories are unknown.
➢ Association: - Identifying the links and relationships among the data
in a single data collection.
• Semi-Supervised Learning: - Given that it works with both labelled and
unlabelled data, semi-supervised learning is a combination of the
supervised and unsupervised approaches outlined above. As a result, it falls
in the middle between learning "without supervision" and learning "with
supervision". Semi-supervised learning is helpful when there are many
unlabelled data sets and few labelled ones in the actual world. In the end,
a semi-supervised learning model should be able to predict outcomes that
are superior than those obtained by simply using the model's labelled data
alone. Machine translation, fraud detection, data labelling, and text
categorization are a few examples of application domains where semi-
supervised learning is applied.
• Reinforcement Learning: - The agents learn in this type of learning by
receiving rewards. Even though there are start and finish points, the agent's
objective is to use the most direct and efficient route to the destination.
Positive awards are offered to the agent when she follows the proper
procedures. But choosing the wrong path results in undesirable outcomes.
Learning happens while working toward a goal.
Analyses of Frequently Used Machine Learning Algorithms
• Decision Tree Algorithm: - The decision tree method is a classic
technique that is frequently used for machine learning. Its basic
operation is to begin processing data information from the collection
instance's root node and proceed until it reaches the point where the
nodes come together. Scientific division of real-world examples. The
decision number algorithm will continue to separate branches to make
data analysis easier while also trimming the branches to improve the
integrity of the data content. The algorithm falls under the top-down
algorithm category from a calculation standpoint. The node is expanded
to more than two depending on the node after the content of the node
is examined during the content analysis process for the best attributes.
For instance, you could designate the decision tree with a lot of data
information as the larger tree A when you are evaluating data, and you
could also choose the maximum number of branch splitting. If the upper
limit is set to 5, the larger tree A will stop splitting once it reaches that
value and instead utilise the pruning method to analyse the larger tree
model in order to clean up the data and increase the objectivity of the
data analysis findings.
➢ Root nodes: It is a node with the ability to develop one or more
branches but no existing branches. The dependent variable is
displayed by the root node, along with the variable that will be
utilised for classification.
➢ Interior nodes: - It is a node that can contain two or more
outbound branches in addition to one incoming branch.
➢ Leaf or Terminal nodes: - The nodes which has incoming branch
but not outgoing branch.
 Fig 2: - Observations of last ten days

Weather

Sunny Cloudy Rainy

Yes Yes
Humidity Wind

High Normal Strong Weak

No Yes No Yes
Fig 3: - Decision Tree Diagram
 • Support Vector Machine (SVM): - The SVM algorithm is another popular
method used in machine learning. The algorithm mostly uses the vector
machine approach in the application procedure to finish the established data
analysis work. The SVM algorithm will also assess the data information that
has to be handled automatically in order to improve the data information.
Several sets of analysis samples must be gathered to identify the sample data
for the boundary value in order to increase the scientific rigour of the final
data analysis conclusions. Assuming, for instance, that the data information
to be processed is H (d), processing H (d) starts with central processing of the
data information using SVM technology, allowing for complete dispersion.
The maximum distance of the entire plane is used to calculate the H (d)
plane's boundary. The accuracy of data processing is improved by analysing
the vector content of the H (d) plane to produce the output vector.

Fig 4: -Diagram of Support Vector Machine

• Regression Analysis: - With the use of numerous machine learning

techniques, regression analysis enables the prediction of a continuous (y)
result variable based on the value of one or more (x) predictor variables.
Figure 5 provides an illustration of how classification differs from regression
modelling. The two categories of machine learning algorithms frequently
have some overlaps. Regression models are currently widely utilised in many
different industries, such as financial forecasting or prediction, cost
estimation, trend analysis, marketing, time series estimation, medication
response modelling, and many more. Linear, polynomial, lasso, and ridge
regression are a few of the well-known forms of regression algorithms, and
they are briefly presented in the sections that follow: -
➢ Simple and Multiple Linear regression: - This is a well-known
regression strategy as well as one of the most well-liked ML modelling
techniques. The dependent variable is continuous, the independent
variable(s) may be continuous or discrete, and the regression line's
shape is linear in this method. The best foot straight line is used in
linear regression to establish a relationship between the dependent
variable (Y) and one or more independent variables (X), also known as
the regression line. The following equations give it its definition:

This is one of the most well-liked regression approaches as well as a

common ML modelling tool. In this method, the dependent variable is
continuous, the independent variable(s) may be continuous or
discrete, and the regression line's shape is linear. Using the best ft
straight line, linear regression establishes a link between the
dependent variable (Y) and one or more independent variables (X)
(also known as the regression line). The following equations give the
definition for it: -

Fig 5: -Classification vs Regression

->The dotted line in classification models the linear boundary separating the two
classes, whereas in regression it models the linear relationship between the two
variables.
➢ Polynomial Regression: - In a particular type of regression analysis
known as polynomial regression, the connection between the
independent variable x and the dependent variable y is not linear but
rather polynomial to the degree of nth in x. The polynomial regression
of degree 1 equation, which is defined as follows, is also derived from
the linear regression equation:

Here, x is an independent/input variable and y is the predicted/target

output. B0, B1..., Bn are the regression coefficients. To put it simply, if
data are not distributed linearly but rather are polynomials of the nth
degree, then polynomial regression is used to produce the required
results.
➢ LASSO and ridge regression: -Due to their capacity to reduce over-
fitting and decrease model complexity, LASSO and Ridge regression
are well known as potent techniques that are frequently employed for
developing learning models in the presence of numerous features. The
regression model known as LASSO (least absolute shrinkage and
selection operator) employs the shrinkage-based L1 regularisation
strategy, which penalises "absolute value of magnitude of
coefficients" (L1 penalty). The coefficients seem to be rendered to
zero by LASSO as a result. In order to minimise the prediction error for
a quantitative response variable, LASSO regression searches for the
subset of predictors. Ridge regression, on the other hand, employs L2
regularisation, or the "squared magnitude of coefficients" (L2
penalty). Because of this, ridge regression always produces a non-
sparse solution and compels the weights to be modest. In general,
LASSO regression is beneficial for obtaining a subset of predictors by
removing less significant characteristics, and ridge regression is useful
when a data set has "multicollinearity," which refers to the predictors
that relate to one another.
• K-nearest neighbours(K-NN): -The k-nearest neighbour algorithm, created
by Fix and Hodges in 1951, is predicated on the idea that variables that are
nearby to one another belong to the same class. The basic goal is to classify
newly received data in line with previously classed data. Testing samples
refer to unclassified data, and learning samples refer to material that has
already been classified. The K-NN method selects the k learning instances
 that are closest to the test sample after calculating the distance between
them. The test sample is deemed to belong to that class by using the majority
of the chosen k samples to identify its class (Ozkan,2013).
The following equation determines the separation of the data: -

The K value must be chosen as an odd integer in order to avoid equality

since it will be verified first with the fresh incoming data. According to
Klnc, Boranda, Yücalar, Tunal, Imşek and Zçift (2016), the Cosinus,
Euclidean, and Manhattan distances are used to calculate distance. The
success rate of K-NN classification is also rising when there is a large
amount of learning data present. Additionally, outcomes in noisy data
are extremely effective. There are drawbacks in addition to these
advantages, though. For instance, it is not known with certainty which
distance measurement is utilised when calculating the distance, and it
takes too long to calculate the measurement of the distance between
the test sample and the learning sample (Zkan,2013).

Fig 6: - K-nearest neighbour diagram.

• Artificial Neural Network (ANN): -The term "artificial neural network"
describes a system that replicates the way humans transmit information by
separating different types of data into separate neurons and linking those
neurons via the Internet to perform complicated memory functions. The
artificial neural network method, however, is built on this developing data
analysis process. Each digital unit among the distinguished neurons has a
high level of veracity, and the data can finish the process of external output.
It behaves just like a human body that advances, pauses, and then runs. The
 data information supplied in the artificial neural network algorithm has a
variety of application features, and the relevant analysis process can be
accomplished in accordance with actual needs. Multilayer forward neural
networks (MLFN), self-organizing neural networks (SOM), and artificial
generalised neurons (ART) are currently some of the artificial neural
networks that are frequently utilised [5]. We can pre-set the weighting
coefficient and then pre-set the output threshold in order to simplify the
analysis and calculation of the data. The output of a specific number to the
outside when the calculated total above this threshold enhances the
orderliness of the complete numerical analysis process.

Fig 7: -Artificial Neural Network (ANN)

• Natural Language Processing (NLP): -The study of how computer systems
interact with human (natural) languages, and more specifically, how
computers are programmed to process and analyse enormous amounts of
data from natural languages, is called natural language processing (NLP), and
it is a subfield of computer science and artificial intelligence. It is a subfield
of machine learning that entails performing a predictive analysis on any text.
The Python programming language is compatible with the open-source
machine learning toolkit Scikit-learn. Natural language processing (NLP), a
branch of computer science and artificial intelligence, is the study of how
computer systems interact with human (natural) languages, and more
particularly, how computers are taught to process and analyse massive
volumes of data from natural languages. It is a branch of machine learning
that comprises conducting a predictive analysis on any text. The open-source
machine learning toolkit Scikit-learn is compatible with the Python
programming language.
The following steps are important in NLP: -
➢ Import the dataset at python coding platform such as Jupyter
Notebook, Google Colab .....etc.
➢ Text cleaning or Pre-processing
➔ Eliminate Punctions and Numbers: -Numbers and Punctuation
are useless for processing the given text. If they are included, all
they do is increase the amount of a word list that we produce in
the last step. In addition, they will make the algorithm less
efficient.
➔ Stemming: - To take the root words only.
Awake

Awoke Awoken Awaking Awakes

➔ Convert every word into a lowercase: -Words in a different case,

like "good" and "GOOD," are worthless, for example.
➢ Tokenization: -Tokenization is the process of dividing unstructured
data and natural language text into units of data that can be regarded
as discrete pieces. One can directly utilise a document's token
occurrences as a vector to represent the document.
➢ Stop words: - It is used to eliminate unimportant words, allowing
applications to focus on the important words instead.
➢ Bag of word (BOW): -Bag of words is a text modelling method for
Natural Language Processing. Technically speaking, it is a technique for
feature extraction from text data. This method of extracting features
from documents is straightforward and adaptable. A textual
illustration of word occurrences in a document is called a "bag of
words." We only pay attention to word counts and don't pay attention
to the word order or little grammatical distinctions. It is referred to as
a "bag" of words since any information regarding the arrangement or
structure of the words in the document is ignored. The model is only
 interested in whether recognised words appear in the document, not
where in the document they appear.
➢ Separating the corpus in the Test and Training set: -We require class
train test split from sklearn.cross validation for this. Split can be built
in a 70/30, 70/30, 80/20, 85/15, or 75/25 ratio. Through "test size," I
have selected 75/25 in this instance. Bag and y can have values of 0 or
1. (either in a good or bad way).
➢ Use a predictive model (Here we will use RandomForest Classifier): -
• We will import the RandomForestClassifier module from the
sklearn.ensemble package since RandomForest Classifier is an ensemble
model (meaning it is built of several trees).
• using the "entropy" criterion and 501 trees, or "n estimators."
• The properties X train and y train are used in the. Fit() function to fit
the model.
➢ Predicting Final Results using the X test attribute and the. predict()
method.
➢ We must obtain the confusion matrix in order to understand the
accuracy of the accuracy result. The confusion matrix is a two-by-two
matrix.
• True Positive: -It is a measurement of how many positive things in
the real world have received the proper attention.
• True Negative: -The proportion of true positives that cannot be
accurately determined is calculated.
• False Positive: -It is a measurement of how many actual negatives
are sufficiently acknowledged.
• False Negative: -The number of negatives that are incorrectly
identified in everyday situations is calculated.
Conclusion
We have undertaken a thorough survey of machine learning methods for
intelligent data analysis and applications in this work. We have briefly covered
how different machine learning techniques can be applied to find solutions to a
range of real-world problems in line with our goal. The quality of the data and the
effectiveness of the learning algorithms both have a role in the success of a
machine learning model. The complex learning algorithms must then be taught
using the real-world data and information pertaining to the target application that
have been gathered before the system can support intelligent decision-making.
To demonstrate how machine learning techniques can be applied to a variety of
real-world problems, we also reviewed a few common application areas. The
difficulties encountered, prospective areas for future research, and directions in
the field have all been reviewed and summarised. As a result, the issues that are
highlighted present attractive areas for future study, which must be addressed
with practical solutions for a variety of application domains. Overall, we think that
from a technological standpoint, our research on machine learning-based
solutions points in a promising path and can serve as a reference manual for
further study and applications for professionals in academia and industry as well
as for decision-makers.
References
[1] Li Kanghua, Jiang Shan. Machine Learning and Cultural Production Reform——Based on the
Perspective of the Development of AI Technology [J]. Journal of Xiangtan University (Philosophy and
Social Sciences), 2020, 44 (01): 74-79.

[2] Jiang Na, Yang Haiyan, Gu Qingchuan, Huang Jiya. Machine learning and its algorithm and
development analysis [J]. Information and Computer Science (Theoretical Edition), 2019 (01): 83-84 +
87.

[3] Li Zhiwei. Development of machine learning and several learning methods [J]. Industry and Science
Forum, 2018, 15 (10): 198-199.

[4] Zhang Run, Wang Yongbin. Research on machine learning and its algorithm and development [J].
Journal of Communication University of China (Natural Science Edition), 2018, 23 (02): 10-18 + 24.

[5] Zhang Changshui. Research on the development of machine learning and data mining [C].. 2010-
2011 Development Report of Control Science and Engineering Discipline.: Chinese Society of
Automation, 2018: 82-89 + 223

[6] https://www.javatpoint.com/nlp-analysis-of-restaurant-reviews

[7]https://www.researchgate.net/publication/341875705_Research_on_Machine_Learning_and_Its_
Algorithms_and_Development

[8] https://link.springer.com/article/10.1007/s42979-021-00592-x