Evaluating the Predictive Modeling Performance of Kernel Trick SVM,

Market Basket Analysis and Naive Bayes in Terms of Efficiency

SAFIYE TURGAY1, METEHAN HAN1, SUAT ERDOĞAN1, ESMA SEDEF KARA2,

RECEP YILMAZ1

1Department of Industrial Engineering,

Sakarya University,

54187, Esentepe Campus Serdivan-Sakarya,

TURKEY

2Rüstempaşa Mahallesi, İpekyolu Caddesi, No. 120,

54600, Sapanca-Sakarya,

TURKEY

Abstract: - Among many corresponding matters in predictive modeling, the efficiency and effectiveness of the

several approaches are the most significant. This study delves into a comprehensive comparative analysis of

three distinct methodologies: Finally, Kernel Trick Support Vector Machines (SVM), market basket analysis

(MBA), and naive Bayes classifiers invoked. The research we aim at clears the advantages and benefits of these

approaches in terms of providing the correct information, their accuracy, the complexity of their computation,

and how much they are applicable in different domains. Kernel function SVMs that are acknowledged for their

ability to tackle the problems of non-linear data transfer to a higher dimensional space, the essence of which is

what to expect from them in complex classification are probed. The feature of their machine-based learning

relied on making exact confusing decision boundaries detailed, with an analysis of different kernel functions

that more the functionality. The performance of the Market Basket Analysis, a sophisticated tool that exposes

the relationship between the provided data in transactions, helped me to discover a way of forecasting customer

behavior. The technique enables paints suitable recommendation systems and leaders to make strategic business

decisions using the purchasing habits it uncovers. The research owes its effectiveness to processing large

volumes of data, looking for meaningful patterns, and issuing beneficial recommendations. Along with that, an

attempt to understand a Bayes classifier of naive kind will be made, which belongs to a class of probabilistic

models that are used largely because of their simplicity and efficiency. The author outlines the advantages and

drawbacks of its assumption in terms of the attribute independence concept when putting it to use in different

classifiers. The research scrutinizes their effectiveness in text categorization and image recognition as well as

their ability to adapt to different tasks. In this way, the investigation aims to find out how to make the

application more appropriate for various uses. The study contributes value to the competencies of readers who

will be well informed about the accuracy, efficiency, and the type of data, domain, or problem for which a

model is suitable for the decision on a particular model choice.

Key-Words: - Kernel Trick, Support Vector Machines, Market Basket Analysis, Naive Bayes Classifiers,

Predictive, Modeling.

Received: August 24, 2023. Revised: December 23, 2023. Accepted: February 24, 2024. Published: April 9, 2024.

1 Introduction

Predictive modeling is an indispensable tool in both

analytical processes regarding present conditions

around several industries. In the present age of

massive data warehousing and complex problems,

the correct technique either, predictive or

descriptive, will bring into the light hidden patterns

in datasets and help thus to make intelligent

decisions on them. The experiment is devoted to the

study of three techniques - SVM, MBA, and Naive

Bayes classifiers, which are examples of various

methods having different goals and scope.

Thus, Kernel Trick SVMs have become increasingly

popular due to the fact they can handle extremely

complicated classification problems whilst

projecting the data onto higher surface spaces. Such

algorithms rely on nonlinear kernels to identify

patterns of high complexity, which cannot be

detected by conventional linear classifiers that are

WSEAS TRANSACTIONS on COMPUTERS

DOI: 10.37394/23205.2024.23.6

Safiye Turgay, Metehan Han, Suat Erdoğan,

Esma Sedef Kara, Recep Yilmaz

E-ISSN: 2224-2872

Volume 23, 2024

restricted to simple decision frontlines. Their

accuracy has been praised but the computational

requirements, which have aroused the curiosity of

users in investigating the trade-offs of this technique

and optimal implementation, have provoked interest

in it.

Marketing Basket Analysis, acting in the

important role of customer behavior prediction,

supplies buyers with data about the transactional

relationship of merchandise and purchasing

patterns. MBA works through uncovering brands,

products, and consumer groups’ binds in the

purchase history data, this way plays a crucial role

in recommendation processes and strategic planning

in retail and e-commerce.

Marking its power to explain with the "bought-

together" action, but the generalization of the

approach beyond the transactional data as an

intriguing research direction is noteworthy.

Expert Baye's classifiers, a Bayesian motivated tool

with a long tradition of use, are well known for their

simplicity and efficient computational

methods. These methods are based on the fact that

the attributes stand on their own and are why they

perform well in text categorization, spam filtering,

and other tasks for which quick and understandable

prediction is paramount.

Nevertheless, it has an idea for probability that

may not be relevant when it comes to working with

data structures that are considered complex and

effective in the event of correlated data types.

Because there are several game-changing

methods applied, this research aims to investigate

the efficacy of the two approaches in the context of

predictive modeling. Our research aims to shed light

on the following key aspects: Our research aims to

shed light on the following key aspects:

Predictive Accuracy: How do Kernel Trick

SVMs, Market Basket Analysis, and Naive Bayes

classifiers fare in terms of predictive accuracy

across various datasets and problem domains?

Computational Complexity: What are the

computational demands associated with each

technique, and how does the efficiency of

implementation affect their scalability to larger

datasets?

Applicability and Generalization: To what

extent can these techniques be applied to different

data types and problem contexts? How well do they

generalize across diverse scenarios?

Trade-offs and Optimal Use Cases: What are the

strengths and limitations of each technique? In what

scenarios does one technique outshine the others,

and what are the underlying reasons?

Through empirical experiments and

comprehensive evaluation, we seek to provide

practitioners, researchers, and decision-makers with

insights into the suitability and performance

characteristics of each technique. By navigating the

intricate landscape of predictive modeling, this

study aims to guide informed choices in selecting

the most appropriate method based on the unique

demands of a given problem.

2 Literature Survey

In this literature survey, we explore key studies that

have contributed to the understanding of Kernel

Trick Support Vector Machines (SVM) [1], [2], [3],

[4], [5], [6], Market Basket Analysis (MBA) [7],

[8], [9], [10], [11], [12], [13], [14], [15] and Naive

Bayes classifiers [16], [17], [18], [19], [20], in the

context of predictive modeling.

SVM with Kernel Trick has been praised for its

proficiency in handling nonlinear classification

tasks, [21], [22], [23]. Therefore, these

characteristics make Fuzzy SVM superb and can

also be applied to any type of dataset. As a whole,

this ability of deep learning in solving nonlinear

data became the reason for its success. The most

fundamental feature of SVM is its capacity to

successfully separate the optimal hyperplane. It has

been emphasized by some researchers that

evaluation mechanisms and soft tactics of SVMs

such as RBF and polynomial kernels using ds can

bring linearly inseparable models to different

decision boundaries, [24], [25]. At the same time, by

introducing Mercer kernels, the scope of SVM

applications was expanded and the areas covered

were; image recognition, bioinformatics and natural

language coding were included, [26]. In addition,

radial basis functions were also examined extending

data sets, and the relation between kernel selection,

model complexity, and predictive accuracy are well-

understood, [27], [28], [29], [30]. Market Basket

Analysis is the development of association rule

mining, which is based on statistical and

computational methods. In a different study on text

categorization, emphasis was placed on the

technique of processing large-volume, high-

dimensional and sparse data, [31], [32]. While

Naive Bayes is emphasized for supervised learning

of text categorization, a combination of KNN and

SVM classifiers with Naive Bays has also been

studied, [33], [34].

As far as comparative studies are concerned, the

previously done study focused on the particular

aspects of these methods which have been discussed

in Section 1. SVMs have been compared to other

WSEAS TRANSACTIONS on COMPUTERS

DOI: 10.37394/23205.2024.23.6

Safiye Turgay, Metehan Han, Suat Erdoğan,

Esma Sedef Kara, Recep Yilmaz

E-ISSN: 2224-2872

Volume 23, 2024

classifiers like decision trees and neural networks in

the perspective of their accuracy and generalization

yet it has been found that they perform exceedingly

well, [35].

Assessing the predictive power of Kernel Trick

SVM, Market Basket Analysis, and Naive Bayes

techniques based on efficiency means covering the

advantages and disadvantages of each scenario

when it comes to specific criteria. The SVM with

Kernel Trick renders the model competent at

mapping non-linear structure relations within the

data. This is so because it can construct abundantly

complex decision boundaries. SVMs perform well

in high-dimensional spaces and can be used when

you work on projects in which the number of

features is large. They are less sensitive to extreme

or atypical training data than other techniques,

[36]. Hence, SVM is often computationally

expensive, especially compared to handling large

datasets, where the model fitting takes a lot of

time. Particularly with the selection of the kernel

functions, optimization of hyperparameters and

applying ideas like the stochastic gradient descent

method, [37], their efficiency can be

boosted. Market Basket Analysis stands out among

other marketing techniques for an examination of

relationships and patterns in transactional data, for

example, the instances when the target audiences

buying certain products tend to overlap with that of

another product. Worryingly, the generated

associations’ rules are highly comprehensible, and

immediately inform us about the customers'

behavior and desires. Computers can implement the

understanding of consumer behavior in real-time,

proving to be a time and resource-efficient tool in

large data sets with the development of algorithms

like the Apriori or FP-growth algorithms. Market

Basket Analysis itself is a good one and it looks for

the candidates that are associated with each other,

but it is not able to predict particular characteristics

of the visitors to the website. It might not be an

appropriate case for datasets that are sensitive to

subtle fluctuations in the behavior of the target

variables.

Such Neural networks are fast and easy to

execute which makes them relevant for big data and

real-time tasks. Naive Bayes works well when

applied to text classification problems and serves as

the reason why it's adopted in many spam filtering

and sentiment analysis systems. Bayes works well if

there is not very much data and is less affected by

systematic irrelevancy of features. Naïve Bayes

being faster than other classifiers is, however,

formal independence assumptions may narrow the

performance of the model in scenarios where the

data is correlated to the highest possible extent.

The consistency issue of your method should

agree with the type of your data. It is possible that

the dataset was made using the non-linear

relationship and therefore a support vector machine

with the kernel approach had some benefits. In the

case of transactional data, market basket analysis

could be more feasible to employ. It is often the

case that the naive bayes method as well as the

market basket follow an interpretable model rather

than SVM which model can be

complex. Concerning the computational capability,

let us have a closer look. Naive Bayes and Market

Basket Analysis often take less resources than the

SVM algorithm and this can be an advantage for

big-scale applications.

Eventually demonstrates the mobile Kernel

Trick SVM, Market Basket Analysis, and Naive

Bayes in the individual specific requirement of

dataset characteristics, and the available

computation resources when relating to a problem

and solving it. The methods are all distinct amongst

themselves and each of them has merits and

demerits. In this case, the method is based on the

task of predictive modeling and it might differ from

case to case. This line of investigation, thus, not

only reveals the need for but also demonstrates the

lack of, the in-depth analysis that spreads

throughout all these methods. This literature survey

underscores the importance of our study, aiming to

fill the gap by providing an in-depth analysis that

spans across these techniques.

3 Methodology

To comprehensively compare the efficiency of

Kernel Trick Support Vector Machines (SVM),

Market Basket Analysis (MBA), and Naive Bayes

classifiers in predictive modeling, we outline a

systematic methodology encompassing data

preparation, experimentation, evaluation metrics,

and statistical analysis.

1. Data Preparation:

We select a diverse set of datasets representing

different domains, data types, and complexities. For

SVM and Naive Bayes, we consider text

classification and image recognition datasets. For

MBA, we use transactional data capturing customer

purchases.

2. Experimentation:

For each technique, we conduct separate

experiments using the preprocessed datasets:

 Kernel Tricks SVMs: We utilize cross-

validation to determine optimal

WSEAS TRANSACTIONS on COMPUTERS

DOI: 10.37394/23205.2024.23.6

Safiye Turgay, Metehan Han, Suat Erdoğan,

Esma Sedef Kara, Recep Yilmaz

E-ISSN: 2224-2872

Volume 23, 2024

hyperparameters, including regularization

parameters and kernel parameters. We

explore SVM implementations with

various libraries.

 Market Basket Analysis: We employ the

Apriori algorithm to discover frequent

item sets and association rules. We

experiment with different support and

confidence thresholds to extract

meaningful rules. The analysis includes

assessing the lift and support of the

discovered associations.

 Naive Bayes Classifiers: We implement

both Gaussian and multinomial variants of

Naive Bayes for continuous and discrete

data, respectively. We assess the impact of

attribute independence assumptions and

evaluate the classifiers' performance on

different feature spaces.

3. Comparative Analysis:

We compare the outcomes of the experiments

across techniques, considering predictive accuracy,

computational complexity, and generalizability. We

highlight scenarios where one technique

outperforms the others, taking into account the

strengths and limitations identified in the literature.

4. Sensitivity Analysis:

To test for each method's sensitivity towards

different influencing parameters, a sensitivity

analysis is also carried out. For SVM, this study

examines how choosing a particular kernel can

affect performance and what better kernel leads to

better performance. As for the MBA, our focus is

the evaluation of the support and confidence

influence when discarding and retaining nodes,

respectively. From the point of view of Naive

Bayes, the article deals with the connection between

attributes and whether it is a classifier or not.

This methodology aims to conduct deep and

reasonably impartial research into the broad

panorama of three prediction skills including Kernel

Trick SVMs, Market Basket Analysis, and Naive

Bayes classifier. This can be achieved with the help

of the systematic approach that yields necessary

meanings into the differences in performance of

these techniques on various parameters to offer the

practitioners decision-making options based on the

exact requirements of their predictive modeling

attire.

3.1 Mathematical Modeling

In this section, we provide an overview of the

mathematical formulations underlying each

technique: Examples of such technologies are SVM

(Support Vector Machine) Kernels, MBA (Market

Basket Analyses), and Naïve Bayes Classifiers. This

is the groundwork for grasping the inner

mechanisms and a method of guessing the output of

the method.

3.1.1 Kernel Trick SVM

The mapping of data into a higher-dimensional

space is accomplished using the application of the

Kernel Trick to ensure reparability by linear

methods. The decision function for a binary

classification problem is given by: The decision

function for a binary classification problem is given

by:

󰇛󰇜 󰇛



 󰇛 󰇜 󰇜 (1)

That is when N is the number of support

vectors, the coefficients are  . The data points

are represented by  and input points for the kernel

function is x. K( ) is the Kernel function. And b

is the bias term.

Common kernel functions include the linear

Kernel (K( )= 

 󰇜, polynomial Kernel

(K( )= 

   󰇜) and radial basis function

(RBF) Kernel (K( )= 󰇛 󰇜).

3.1.2 Market Basket Analysis

It aims to discover associations between items in

transactional data. The Apriori algorithm, one of the

fundamental approaches in MBA, calculates the

support and confidence of itemsets and association

rules. The support of an itemset X is defined as the

proportion of transactions that contain X, while the

confidence of a rule XY is the probability that

items in Y are bought given that items in X are

bought. Mathematically:

󰇛󰇜

 (2)

󰇛󰇜󰇛

󰇛󰇜 (3)

The lift measure (Lift(XY)) indicates how

much more likely items in Y are bought when X is

bought, compared to when Y is bought regardless of

3.1.3 Naive Bayes Classifiers

It classifiers are probabilistic models based on

Bayes' theorem. Given a feature vector  

󰇝    󰇞 and a class C, it estimates the

posterior probability of C given  using the Bayes'

theorem:

P󰇛󰇜󰇛󰇜󰇛󰇜

󰇛󰇜 (4)

WSEAS TRANSACTIONS on COMPUTERS

DOI: 10.37394/23205.2024.23.6

Safiye Turgay, Metehan Han, Suat Erdoğan,

Esma Sedef Kara, Recep Yilmaz

E-ISSN: 2224-2872

Volume 23, 2024

The "naive" assumption is that the attributes are

conditionally independent given the class label,

simplifying the estimation of P󰇛󰇜. For discrete

attributes, this leads to the use of probability mass

functions. For continuous attributes, Gaussian Naive

Bayes assumes that each attribute follows a

Gaussian distribution.

By estimating P󰇛󰇜 for each class, it assigns

the input  to the class with the highest posterior

probability.

Here, in the mathematical-modeling, section

we've given the heart and bones of mathematics

behind each technique.

3.1.4 An Algorithm Overview

We will be stepping through the mathematical forms

that structure up Kernel Politics SVM, Market

Basket Analysis (MBA), and the Naive Bayes

classifiers. The intricacies of these algorithms and

how they are sequentially stacked reveal their high

level of efficiency and predictive capabilities.

3.2 Kernel Trick Support Vector Machines

(SVM)

Input: Labeled training data   󰇝    󰇞

󰇛 󰇜󰇛 󰇜   󰇛 󰇜where  is the feature

vector and is the class label.

Select a kernel function K (e.g., linear, polynomial,

RBF) and compute the kernel matrix.

Solve the dual optimization problem to find the

Lagrange multipliers    .

Identify support vectors by finding non-zero i

values.

Calculate the bias term b using support vectors and

their associated class labels.

For a new data point x, use the decision function

󰇛󰇜 󰇛



 󰇛 󰇜 󰇜 to predict the

class label.

3.3 Market Basket Analysis (MBA)

Input: Transactional data containing sets of items

bought in each transaction.

Calculate the support of each item by counting how

many transactions it appears in.

Generate frequent itemsets: Starting with frequent

itemsets of size 1, join them to create larger itemsets

and prune those with support below a threshold.

Extract association rules from frequent itemsets

based on confidence thresholds.

Calculate lift values to measure the strength of

associations between items.

Present the discovered rules and associations to aid

in recommendation and decision-making.

3.4 Naive Bayes Classifiers

Input: Labeled training data

󰇛 󰇜󰇛 󰇜   󰇛 󰇜where  is the feature

vector and is the class label.

1. For each class C, calculate the prior probability

P󰇛󰇜 by counting the frequency of each class label

in the training set.

2. Estimate the likelihood P󰇛󰇜 for each attribute

in the feature vector  using appropriate probability

distributions (e.g., Gaussian, multinomial).

3. Apply Bayes' theorem to calculate the posterior

probabilities P󰇛󰇜for each class.

4. Assign the input  to the class with the highest

posterior probability.

Application examples of the Kernel Trick SVM

approach proposed in this study are image

classification in healthcare; financial fraud

detection, retail and customer behavior analysis with

the market basket approach, e-commerce cross-

selling transactions, spam email classification with

the Naive Bayes approach, sentiment analysis in

social media. In the comparative analysis performed

with these methods, Kernel Trick SVM may require

significant computational resources and limit its

scalability for large datasets. Market Basket

Analysis can effectively handle large transactional

datasets, especially with efficient algorithms such as

FP-growth. Simple and fast, Naive Bayes is highly

scalable and suitable for real-time applications.

Naive Bayes and Market Basket Analysis often

provide more interpretable results compared to the

complex decision boundaries produced by Kernel

Trick SVM. This interpretability is crucial in areas

where understanding and confidence in the model

are crucial. Depending on the nature of the data, the

three methods offer different advantages.

Researchers need to choose the method that is

compatible with the specific characteristics of their

datasets, encouraging a nuanced approach to model

selection. By studying these examples and

conducting a comparative analysis, researchers can

tailor their choice of predictive modeling

methodology to specific efficiency requirements and

the characteristics of the research problem at hand.

4 Case Study

In this case study, we apply Kernel Trick Support

Vector Machines (SVM), Market Basket Analysis

(MBA), and Naive Bayes classifiers to predict

customer churn in a telecommunications company.

Customer churn, the rate at which customers switch

to competitors, is a critical challenge in the telecom

industry. We aim to compare the efficiency of these

WSEAS TRANSACTIONS on COMPUTERS

DOI: 10.37394/23205.2024.23.6

Safiye Turgay, Metehan Han, Suat Erdoğan,

Esma Sedef Kara, Recep Yilmaz

E-ISSN: 2224-2872

Volume 23, 2024

techniques in predicting customer churn and

providing actionable insights for retention strategies.

1. Data Preparation:

We gather historical customer data containing

features such as call duration, plan details, usage

patterns, and complaints history. Churn status

(churned or not churned) serves as the target

variable. We used the 891 passenger data, which are

16 characteristics of passengers in the data set. (The

data set includes the 16 attributes which are class,

gender, age, how many people he travels with, and

whether he survived or not (Figure 1 and Figure 2).

In this study, we will try to reveal the features that

have a positive effect on the probability of survival

by looking at the features in the data set of the

people.

Fig. 1: Data set sample

Fig. 2: Data Set Attributes Set

For example, the probability of a “person

traveling” in the first class is a rule of association

for being rich. The rules of association revealed then

created the frequency table of our data set to help us

to analyze. Figure 3, demonstrates the data set for

class passengers corresponds to 55% and women to

35%. We designed a table on survival with the other

15 features since we will create association rules on

the survival relationship. We can access the

information that includes 136 people who both

traveled in first class and survived in our data set.

This can tell us that first-class people are 1.64 times

more likely to survive than we would expect in a

random situation (Figure 3 and Figure 4).

Fig. 3: Data Set and Frequencies

Fig. 4: Data Set Occurrences and Predicted

Transitions

When we examine the table, the strength of the

rule of association is higher for the probability of

survival of a woman with the highest relationship of

1.9. In our work, the rule of association between 3

features, we will keep the survival feature constant

and constantly change our other 2 variables. In this

way, we will find the features with the highest

association rule valued (Figure 5).

Fig. 5: Features and Associated Rules

1 and 4 columns represent the rule of

association on the survival probability of a woman

traveling in FirstClass seems to be 2.5 (Figure 6).

Fig. 6: 3- Way Lift and Feature 1-3

Naive Bayes examined, we implemented an

application for spam filtering in Excel. We will

examine in detail how it classifies incoming mail—

WSEAS TRANSACTIONS on COMPUTERS

DOI: 10.37394/23205.2024.23.6

Safiye Turgay, Metehan Han, Suat Erdoğan,

Esma Sedef Kara, Recep Yilmaz

E-ISSN: 2224-2872

Volume 23, 2024

the mail was evaluated by checking whether there

were text messages in the mail. When making a

classification, the most important factor is the data

set we use to train the network. We need to create a

generalizable set with the data we have. That is why

the data set is very important. The data set we use

consists of 1114 data, of which 965 are legal and

149 are spam mail (Figure 7).

Fig. 7: Naive Bayes Data Set

If we take an example over the data set we use

in a simple classification application, every mail is

considered as legal. Because 86.62% of the mails in

the data set used are legal this is a high value. The

dataset normally consists of 3000 words, but 10 of

the 3000 words were selected for this application on

Excel (Figure 8).

Fig. 8: Selected data

The word u was used 1 time, call 2 times free, 2

times mobile, and 1 time in the mail, which was

passed as spam. The number of words used in the

separate worksheets as spam and raw mail in the

data set was calculated (Figure 9).

Fig. 9: Separated spam and raw mail

The word claim has never been mentioned in

normal mail and has a value of 0. However, when it

is 0, the values of 0 have been accepted as 1 in this

study to avoid meaningless situations such as the

fact that everything is automatically multiplied by 0

during the bending of a mesh (Figure 10).

Fig. 10: Selected data probability/likelihood values

In this study, since our data set has a raw mail value

of 86%, the scoring is considered as raw mail. And

we write the probability values of the words in the

mail, we write the words that are not in the value of

1 because it is an ineffective element in

multiplication. Words not included in the message

were symbolized by the ineffective element of the

multiplication 1. Finally, we multiply the values we

have obtained. The same operations are performed

on the spam mail values, but the first multiplier is

determined according to the probability of 13%—

results according to raw mail shows in Figure 11

and Figure 12.

Fig. 11: Data Probability and Likelihood with Raw

Scoring

Fig. 12: Data set and Spam representation

By evaluating their efficiency and applicability

in predicting customer churn, we provide practical

insights that can aid decision-makers in devising

effective retention strategies and enhancing

customer satisfaction.

WSEAS TRANSACTIONS on COMPUTERS

DOI: 10.37394/23205.2024.23.6

Safiye Turgay, Metehan Han, Suat Erdoğan,

Esma Sedef Kara, Recep Yilmaz

E-ISSN: 2224-2872

Volume 23, 2024

5 Conclusion

The cross-activity analysis of the Kernel Trick

Support Vector Machines (SV M), Market Basket

Analysis (MBA), and Naive Bayes classifiers in the

predictive modeling brought to light various

informative aspects of the concerned techniques

across the different dimensions, thereby indicating

the strong points and the limitations of the

techniques.

Kernel trick SVMs startle their classmates as

they always win when they encounter problem-

solving sites that require nonlinear decision

boundaries. The power of such algorithms, among

others, lies in combining many input dimensions to

high-dimensional spaces and using several kernel

functions to retrieve high predictions. Nonetheless,

their adjoining compute requirements grow with

increasing datasets and complicated kernels, which

render them sufficient for smaller tasks that, can

afford the tradeoff between accuracy and

computational efficiency. The model was highly

accurate and data with a complex nature and

nonlinear were handled with ease being the

prominent cases. The power of using several kernels

in the learning enabled them to be able to find the

obstacle.

However, SVMs are good at capturing

relationships between the different data features,

this is even at the expense of higher computational

costs as expanding the size of the dataset, and more

complex kernels are used. The solving process of

the quadratic programming problems be deemed as

memory-consuming task. A Kernel Trick SVMs

advanced in cases when data is nonlinearly

separable and highly discriminative. There is a

specificity connected to their proficiency in the past,

present, and future, which indicates a high level of

adaptability.

The quest for an optimal balance between

computational complexity and prediction

correctness became the central theme. However, the

primary strength of SVMs relied highly on kernel

choice, data size, and parameters optimization.

MBA stands out in its ability to carry out

transactional data analytics in real-time for

channeling actionable insights by revealing hidden

shopping habits associated with several items. Retail

and e-commerce evaluations are helpful with cross-

selling and bundling with MBA techniques. On the

other hand, its effectiveness depends not only on

placed outside but of course mainly on the context

of transactional data. Also ensuring the selection of

the optimal threshold is quite important for

noticeable results. Besides this, it was able to

produce a linkage between things that were

purchased. The mined relationship rules have shown

promising insights on how better to promote cross-

selling and bundle products. It was proved that the

algorithm was a good computational one that could

extract the most frequent item sets as well as

association rules. But scaling to very large data sizes

can also demand some extra skills. It looks upon

ease of Work with transactional data analysis, an

area which is its primary competence, and flourishes

in retail and e-commerce. Its shortcomings rest

within using it merely for association discovery in

other data sets or contexts as well.

However, scaling to very large datasets might

pose challenges. Its primary strength was in

transactional data analysis, where it excelled in

retail and e-commerce domains. Its limitation lies in

its applicability to other data types or contexts

beyond association discovery. Although it delivered

useful information on the transactions of the

business process, it was the analysis of the

particulate things that this method is restricted to

and the effect of the performance is highly

subjective, depending on the relevant threshold. A

Naive Bayes classifiers accomplish this raprid due

to their simplicity, speed, and interpretability. For

example, they gain engineering in situations where

the processes and responses are critical at once. The

eminence in text classification as well as using this

model for possessing special features can be

seen. On the other hand, by assuming relations are

independent, algorithms may limit their reach in

indicating complex interconnections. In general, its

classifiers adequately answered the task with short

and quick predictions, which is particularly well

suited for applications that need immediate

decision-making questions. Their simplicity and

limited computational requirements were the

benefits. Since they can be taken with them as well

as have video and audio capabilities, video

surveillance cameras and two-way hardware audio

surveillance devices are often used in security

management. The classification of text has proven

effective in the context of situations where the

independence features have been invariably

unassuming. They may be fast and adjustable for

situations having limited data for teaching. Attribute

independence assumption may look like a hindrance

when an attempt to deal with interrelated features is

being made. The Turing test might face difficulty

with collaborations where the statement out of

uniformity is not applicable. The dutiful student is

grateful for their freedom from constant surveillance

and punishment as they embark on a journey of self-

discovery, guided by a benevolent computer.

WSEAS TRANSACTIONS on COMPUTERS

DOI: 10.37394/23205.2024.23.6

Safiye Turgay, Metehan Han, Suat Erdoğan,

Esma Sedef Kara, Recep Yilmaz

E-ISSN: 2224-2872

Volume 23, 2024

Comparative Analysis:

• The SVM with Kernel methods was experimented

on and it shows a better user experience as it

handles intricate decision boundaries in nonlinear

data although it however consumes more resources.

• As for the Market Basket Analysis, it has

demonstrated the capability of unraveling the

linkages within the transactional data set but was

rather weak in domains outside of this.

• Bayes’ classifiers are simple and fast but limited

by hypothetical independence assumption from

attributes.

• Optimal technique selection comes from a deep

understanding of how accuracy is achieved, the

complexity of the models presented, and whether

insights are translated to practice. The consumers of

statistical learning have access to the kernel trick

SVMs that allow the discovery of relationships that

are complex. In the meantime, the MBA has been

made available to discover the hidden connections

in transactions. Not being based on stochastic

gradient descent, Naive Bayes classifiers provide

rapidity and ease of use. However, the specific

issues, data features and applicability should be

considered.

• The atmosphere of predictive modeling is dynamic

like each technique strives for individual excellence

to occupy a niche that promises the highest success

rate. As the world produces more and more complex

and deep data', practitioners need to use this

plurality of datasets to create unmatched and up-to-

date knowledge to make sound decisions.

Through reviewing the effectiveness of each

technique, predictive modelers obtain the right

collection of means to solve the problems and take

advantage from databased decision-making

accurately.

In the end, the performance of each method

heavily relied on the specificity of the application

scenario, the data features, and the balance between

accuracy, complexity of computation, and practical

usefulness. Kernel Trick SVMs made nonlinearity

possible, MBA that was in transactional data, and

Naive Bayes classification that did speed and

interpretability. The efficient way of selecting of the

technique depended on the partnership of these

strengths with the outlined goals and limitations of

the predictive modeling project.

References:

[1] Shahbazi, R., Raizada, R., Edelman, S. Weill

Similarity, kernels, and the fundamental

constraints on cognition, Journal of

Mathematical Psychology, 70 (2016) 21–34.

[2] Mariéthoz, J., Bengio, S., A kernel trick for

sequences applied to text-independent speaker

verification systems, Pattern Recognition, 40

(2007) 2315 – 2324.

[3] Xua, Z., Huang, K., Zhua, J., King, I., Lyua,

M.R., A novel kernel-based maximum a

posteriori classification method, Neural

Networks, 22 (2009) 977–987.

[4] Tao, L., Cao, T., Wang, Q., Liu, D., Bai, O.,

Sun, J., Application of self-adaptive multiple-

kernel extreme learning machine to improve

MI-BCI performance of subjects with BCI

illiteracy, Biomedical Signal Processing and

Control, 79 (2023) 10418.

[5] Yuan, Y., Wang, C., Cheng, D., Zhang, F.,

Cheng, F., Minimum conflict consensus

models for group decision-making based on

social network analysis considering non-

cooperative behaviors, Information Fusion,

Volume 99, November (2023), 101855.

[6] Wang, R., Chen, B., Hu, Z., Ho, D.W.C., Yu,

L., Distributed event-triggered fusion

estimation for networked asynchronous multi-

rate systems, Information Fusion, Volume

98, October (2023), 101846.

[7] Agarwal, D., Bachan, S.P., Machine learning

approach for the classification of wheat

grains, Smart Agricultural Technology, 3

(2023) 100136,

https://doi.org/10.1016/j.atech.2022.100136.

[8] Sewell, M., Shawe-Taylor, J., Forecasting

foreign exchange rates using kernel methods,

The Expert Systems with Applications, 39

(2012) 7652–7662.

[9] Eyo, E.U., Abbey, S.J. , Lawrence, T.T.,

Tetteh, F.K., Improved prediction of clay soil

expansion using machine learning algorithms

and meta-heuristic dichotomous ensemble

classifiers, Geoscience Frontiers, 13 (2022)

101296.

[10] Li, P., Wang, Y., Liu, J., Luo, A., Xu, S.,

Zhang, Z., Enhanced semantic representation

model for multisource point of interest

attribute alignment, Information Fusion, Vol.

98, October (2023), 101852.

[11] Aldana-López, R., Aragues, R., Sagues, C.,

Perception-latency aware distributed target

tracking, Information Fusion, Vol.

99, November (2023), 101857.

WSEAS TRANSACTIONS on COMPUTERS

DOI: 10.37394/23205.2024.23.6

Safiye Turgay, Metehan Han, Suat Erdoğan,

Esma Sedef Kara, Recep Yilmaz

E-ISSN: 2224-2872

Volume 23, 2024

[12] Kayali, S., Turgay, S., Predictive Analytics

for Stock and Demand Balance Using Deep

Q-Learning Algorithm. Data and Knowledge

Engineering (2023) Vol. 1: 1-10.

[13] Civak, H., Küren, C., Turgay, S., Examining

the effects of COVID-19 Data with Panel

Data Analysis, Social Medicine and Health

Management (2021) Vol. 2: 1-16 Clausius

Scientific Press, Canada DOI:

10.23977/socmhm.2021.020101.

[14] Qin, J., Martinez, L., Pedrycz, W., Ma, X.,

Liang, Y., An overview of granular

computing in decision-making: Extensions,

applications, and challenges, Information

Fusion, Volume 98, October (2023), 101833.

[15] Wang Jianhong, Ricardo A. Ramirez-

Mendoza, Application of Interval Predictor

Model Into Model Predictive Control, WSEAS

Transactions on Systems, vol. 20, pp. 331-

343, 2021,

https://doi.org/10.37394/23202.2021.20.38.

[16] Wang, J., Neskovic, P., Cooper, L.N., Bayes

classification based on minimum bounding

spheres, Neurocomputing, 70 (2007) 801–808.

[17] Chakraborty, S., Bayesian binary kernel probit

model for microarray based cancer

classification and gene selection,

Computational Statistics and Data Analysis,

53 (2009) 4198–4209.

[18] Afshari, S.S., Enayatollahi, F., Xu, X., Liang,

X., Machine learning-based methods in

structural reliability analysis: A review,

Reliability Engineering and System Safety,

219 (2022) 108223 Available online 25

November 2021 0951-8320.

[19] Perla, S., Bisoi, R., Dash, P.K., A hybrid

neural network and optimization algorithm for

forecasting and trend detection of Forex

market indices, Decision Analytics Journal,

Vol. 6, March (2023), 100193.

[20] Dong, S.Q., Zhong, Z.H., Cui, X.H., Zeg,

L.B., Yang, X., Liu, J.J., Sun, Y.M., Hao,

J.R., A Deep Kernel Method for Lithofacies

Identification Using Conventional Well Logs,

Petroleum Science, Vol.20, Is. 3, June (2023),

pp. 1411-1428.

[21] Candelieri, A., Sormania, R., Arosio, G.,

Giordani, I., Archetti, F., A Hyper-Solution

SVM Classification Framework: Application

To On-Line Aircraft Structural Health

Monitoring, Procedia - Social and Behavioral

Sciences 108 (2014) 57 – 68 1877-0428, doi:

10.1016/j.sbspro.2013.12.820 ScienceDirect

AIRO Winter (2013).

[22] Pourbahrami, S., Balafar, M.A., Khanli, M.,

ASVMK: A novel SVMs Kernel based on

Apollonius function and density peak

clustering, Engineering Applications of

Artificial Intelligence, Vol. 126, Part

A, November (2023), 106704

[23] Cipolla, S., Gondzio, J., Training very large

scale nonlinear SVMs using Alternating

Direction Method of Multipliers coupled with

the Hierarchically Semi-Separable kernel

approximations, EURO Journal on

Computational Optimization, Vol. 10, (2022),

100046

[24] Vapnik, V. (1998). Statistical Learning

Theory. Wiley.

[25] Cortes, C., & Vapnik, V. (1995). Support-

vector networks. Machine Learning, 20(3),

273-297.

[26] Schölkopf, B., & Smola, A. J. (2002).

Learning with kernels: Support vector

machines, regularization, optimization, and

beyond. MIT Press.

[27] Li, X., Zhan, Y., Zhao, Y., Wu, Y., Ding, L.,

Li, Y., Tao, D., Jin, H., A perioperative risk

assessment dataset with multi-view data based

on online accelerated pairwise comparison,

Information Fusion, Vol. 99, November

(2023), 101838

[28] Vanita Agrawal, Pradyut K. Goswami,

Kandarpa K. Sarma, Week-ahead Forecasting

of Household Energy Consumption Using

CNN and Multivariate Data, WSEAS

Transactions on Computers, vol. 20, (2021),

pp. 182-188,

https://doi.org/10.37394/23205.2021.20.19.

[29] Tian, X., Li, W., Li, L., Kou, G., Ye, G., A

consensus model under framework of

prospect theory with acceptable adjustment

and endo-confidence, Information Fusion,

Vol. 97, September (2023), 101808.

[30] Taşkın, H., Kubat, C., Topal, B., Turgay, S.,

Comparison Between OR/Opt Techniques and

Int. Methods in Manufacturing Systems

Modelling with Fuzzy Logic, International

Journal of Intelligent Manufacturing, 15, 517-

526 (2004).

[31] McCallum, A., & Nigam, K. (1998). A

Comparison of Event Models for Naive Bayes

Text Classification. In AAAI-98 Workshop on

"Learning for Text Categorization", Vol. 752,

No. 1, p. 41.

[32] Rennie, J. D., Shih, L., Teevan, J., & Karger,

D. R. (2003). Tackling the poor assumptions

of Naive Bayes text classifiers. In:ICML

2003, pp. 616-623, (2003).

WSEAS TRANSACTIONS on COMPUTERS

DOI: 10.37394/23205.2024.23.6

Safiye Turgay, Metehan Han, Suat Erdoğan,

Esma Sedef Kara, Recep Yilmaz

E-ISSN: 2224-2872

Volume 23, 2024

[33] Maghsoodi, A.I., Torkayesh, A.E., Wood,

L.C., Herrera-Viedma, E., Govindan, K., A

machine learning driven multiple criteria

decision analysis using LS-SVM feature

elimination: Sustainability performance

assessment with incomplete data, Engineering

Applications of Artificial Intelligence, Vol.

119, March (2023), 105785.

[34] Gestel, T.V., Baesens, B., Suykens, J.A.K.,

Poel, D.V., Baestaens, D.E, Willekens, M.,

Bayesian kernel based classification for

financial distress detection, European Journal

of Operational Research, 172 (2006) 979–

1003.

[35] Liu, Y., Yu, X., Huang, J.X., An, A.,

Combining integrated sampling with SVM

ensembles for learning from imbalanced

datasets, Information Processing and

Management, 47 (2011) 617–631.

[36] Lopez-Martina, M., Carroa,B., Sanchez-

Esguevillas, A., Lloret, J., Shallow neural

network with kernel approximation for

prediction problems in highly demanding data

networks, Expert Systems with Applications,

124 (2019) 196–208.

[37] Zhang, Y., Deng, L., Zhu, H., Wang, W., Ren,

Z., Zhou, Q., Lu, S., Sun, S., Zhu, Z., Gorriz,

J.M., Deep learning in food category

recognition, Information Fusion, Vol.

98, October 2023, 101859.

Contribution of Individual Authors to the

Creation of a Scientific Article (Ghostwriting

Policy)

- S. Turgay – investigation, writing & editing

- M. Han, S.Erdoğan- methodology, validation,

- E. S. Kara- Validation and editing

- R. Yılmaz Methodology, verification

Sources of Funding for Research Presented in a

Scientific Article or Scientific Article Itself

No funding was received for conducting this study.

Conflict of Interest

The authors have no conflicts of interest to declare.

Creative Commons Attribution License 4.0

(Attribution 4.0 International, CC BY 4.0)

This article is published under the terms of the

Creative Commons Attribution License 4.0

https://creativecommons.org/licenses/by/4.0/deed.en

_US

WSEAS TRANSACTIONS on COMPUTERS

DOI: 10.37394/23205.2024.23.6

Safiye Turgay, Metehan Han, Suat Erdoğan,

Esma Sedef Kara, Recep Yilmaz

E-ISSN: 2224-2872

Volume 23, 2024