Journal of Science and Technology

ISSN: 2456-5660 Volume 8, Issue 9 (Sep -2023)

www.jst.org.in DOI:https://doi.org/10.46243/jst.2024.v8.i9.pp1-8

A Method for Vibration Testing Decision Tree-Based

Classification Systems.

To Cite this Article

B.Siva, Ruhi.“ A Method for Vibration Testing Decision Tree-Based Classification Systems”
Journal of Science and Technology, Vol. 08, Issue 09 - Sep 2023, pp1-8

Article Info
Received: 28-08-2023 Revised: 05 -08-2023 Accepted: 22-08-2023 Published: 7-09-2023

Abstract— "Without any intervention from a person, computers are capable of "learning" new
things by analyzing data in various ways (training and testing) and making conclusions. One
application of ML is decision trees. Many diverse disciplines make use of decision tree techniques.
These algorithms have a wide variety of potential applications, including search engines, text
extraction, and companies that provide medical certifications. Decision tree algorithms that are
both accurate and affordable are now at our fingertips. Whenever a choice is necessary, it is critical
to know what the best choice is. We present three decision tree algorithms in this study: ID3, C4.5,
and CART. We use tools like WEKA, ML, and DT.

I. EXPLORING THE DECISION TREE

Classification is the process of assigning things to categories, and it has many different uses.

Fig. 1: Classification of mapping attribute set (X) to its class label (Y)

Decision Diagram
Trunk, branches, and leaves are the typical components of a tree. Decision Trees follow the same
pattern. A tree's trunk, branches, and foliage give this structure its distinctive appearance. Attribute
testing occurs at each leaf node [3, 4]. The test results are shown at the leaf node and continued
down the branch. As its name implies, the root node is the initial node in a tree and acts as the
biological parent to all the other nodes. According to [4], a "node" represents a quality or attribute,
a "branch" a choice or rule, and a "leaf" a result, whether continuous or categorical. Decision trees
provide quick data collecting and accurate conclusion drawing since they are designed around
human thought processes. With the goal of processing a single result at each leaf, we want to build
such a tree for all the data.

CONNECTED RESEARCH ON THE DECISION TREE

 Journal of Science and Technology
ISSN: 2456-5660 Volume 8, Issue 9 (Sep -2023)
www.jst.org.in DOI:https://doi.org/10.46243/jst.2024.v8.i9.pp1-8

Decision Tree is straightforward as it attempts to simulate human decision-making process. Issues

that persist independent of the data type (continuous or discrete). An example of a Decision Tree
is shown here [15].

Fig. 2: Example of Decision Tree on what to do whendifferent situations occur in weather.

Splitting is instantly terminated if any data is deemed useless. Finding specific tests is more
effective than trying to optimize the tree overall. Bear in mind that the data set only gives
categorical information, and that the ID3 technique can only be simulated using the WEKA tool,
when you analyze the properties of Decision Tree. Under ID3's simulation conditions, continuous
data collecting is not possible. A number of parallels exist between CART and C4.5.
features identical to those of ID3. C4.5 and CART are similar in that both may use continuous data sets as
input for simulation purposes [11], but there is one key distinction.

Table-1: Characteristics of DT.

Decision Data Types Numerical Possib
Tree Data le Tool
Splitting
Algorithm Method
CHAID Categorical N/A SPSS
answ
er tree
ID3 Categorical No WEKA
Restriction
C4.5 Categorical, No WEKA
Numerical Restriction
CART Categorical, Binary Splits CART 5.0
Numerical

A One way to organize all the potential outcomes and the actions needed to get to each one is using
a decision tree [12]. One of Decision Tree's strengths is how transparent and honest it is.You also
 Journal of Science and Technology
ISSN: 2456-5660 Volume 8, Issue 9 (Sep -2023)
www.jst.org.in DOI:https://doi.org/10.46243/jst.2024.v8.i9.pp1-8

get to choose the most biased and understandable nature, which is a huge plus. Both its
classification and comprehension are within my capabilities. Able to effortlessly process info that
is either continuous or discrete. The decision tree only needs to be able to segment features and
filter variables [19]. Despite its impact on performance, non-linear makes no adjustments to the
decision tree's parameters.

Part I: Decision-Based Algorithms Determining the "Best" way to split an attribute between two
categories is possible using a decision tree approach. We need a consistent criteria for creating the
splits if we want the partitions at each branch to be as pure as possible.

datasets. The
technique
called
“PRUNNING”,
solves the
problem of
over- filtering
[9].
C5.0 Improved version C5.0 allows to
of the C4.5 whether
estimate
missing values
as a function of
other attributes
or
apportions the
case
statistically
among the
results
[13].
CHAID (CHi- Predates the For a nominal
original
square ID3 scaled variable,
implementation.
Automatic this type of
Interaction decision tree is
Detector) [6] used. The
technique
detects
the
dependen
t
variable from
the
categorized
variables of a
dataset [3, 11].
MARS (multi- Used to find the In order to
best
adaptive split. achieve the
best
 Journal of Science and Technology
ISSN: 2456-5660 Volume 8, Issue 9 (Sep -2023)
www.jst.org.in DOI:https://doi.org/10.46243/jst.2024.v8.i9.pp1-8

Table- 2: Decision tree algorithms regression split, we can use

splines) the regression
tree
based on
Algorithm Classification Description
MARS
[2, 10].
name
CART Uses Gini Index as a By applying
(Classification metric. numeric splitting,
and Regression we can construct
Trees) the tree based on
CART [4].
ID3 (Iterative Uses Entropy function The only concern
Dichotomiser 3) and Information gain with the discrete
as metrics. values. Therefore,
continuous dataset
must be classified
within the discrete
data set [5].
C4.5 The improved version Deals with both
on ID 3 discrete as well as
a continuous
dataset. Also, it
can handle the
incomplete
 Journal of Science and Technology
ISSN: 2456-5660 Volume 8, Issue 9 (Sep -2023)
www.jst.org.in DOI:https://doi.org/10.46243/jst.2024.v8.i9.pp1-8

I. METRICS
Depending on the settings of the splitting attribute, several subsets of the training data are formed.
Until every instance in a subset belongs to the same class in any Decision Tree, the technique
iteratively iterates [6].

Table- 3: Splitting Criteria

Metrics Equation
Informatio
nGain

Gain Ratio Gain Ratio=I(p,n)-E(A)

I(p,n)= Information
before splitting
E(A)= Information after
splitting
Gini Index

The fundamental issue with Information Gain is its bias towards features that include several
variables [6]. An example of unfair data partitioning would be a child node with an unusually large
record count in comparison to the rest. An increased Gain Ratio is favored [7, 12]. When there are
more than two groups in the data, the credibility of the Gini Index is weakened. Here are several
problems with the splitting criteria [15].The fundamental issue with Information Gain is its bias
towards features that include several variables [6]. An example of unfair data partitioning would
be a child node with an unusually large record count in comparison to the rest. An increased Gain
Ratio is favored [7, 12]. When there are more than two groups in the data, the credibility of the
Gini Index is weakened. Here are several problems with the splitting criteria [15].
Information Gain prioritizes multivariate characteristics over univariate ones, which is a big
problem [6]. In an unfair data partition, one of the child nodes contains a disproportionately high
number of records compared to the others. Advantage is given to greater gain ratios [7, 12]. The
Gini Index is rendered useless since more categories are included in the data. The following issues
often arise while trying to divide criteria: [15].
When elements of a set are very closely packed together, we say that the set is exact. Measuring
accurately involves taking an average of the amounts that have been measured and comparing it
to the actual value of the variable. The only way to measure quantities with more than two terms
is to use data points collected from several measurements of the same quantity [13].
 Journal of Science and Technology
ISSN: 2456-5660 Volume 8, Issue 9 (Sep -2023)
www.jst.org.in DOI:https://doi.org/10.46243/jst.2024.v8.i9.pp1-8

(𝑇𝑃 + 𝑇𝑁)
𝐴𝑐𝑐𝑢𝑟𝑎𝑐𝑦 =
(𝑇𝑃 + 𝑇𝑁 + 𝐹𝑃 + 𝐹𝑁)

𝑇𝑃
𝑃𝑟𝑒𝑐i𝑠i𝑜𝑛 =
(𝑇𝑃 +
𝐹𝑃)

TP = True positive, TN = True Negative FP = False Positive, FN = False Negative

Fig. 3: Confusion Matrix sample in Decision Tree.II. DESCRIPTION OF THE DATASET

The car dataset is used in this investigation. Running this dataset through the CART, ID3, and
C4.5 decision tree algorithms. Here is how this dataset is defined.The car database is composed of
two components. Automotive Technology and Popularity. The vehicle's acceptability is affected
by a number of elements, including its purchase price and the cost to operate it. The amount of
trunk space, safety features, expected passenger capacity, number of doors, and overall size of the
passageway all have a role in crashworthiness.Of these, 1728 are examples. Six characteristics It
is useless to have an attribute if it is not present.Excellence in Personality Traits:

Attribute Attribute
Values
buying v-high, high,
med,
low
maint v-high, high,
med,
low
doors 2, 3, 4, 5-more
persons 2, 4, more
lug_boot small, med, big
safety low, med, high

Class Distribution (Number of instances per class):

Class N N [%]
 Journal of Science and Technology
ISSN: 2456-5660 Volume 8, Issue 9 (Sep -2023)
www.jst.org.in DOI:https://doi.org/10.46243/jst.2024.v8.i9.pp1-8

Unacc 1210 70.023%

Acc 384 22.222%
good 69 3.993%
v–good 65 3.762%

II. EXPERIMENT The experiment is being replicated using WEKA. The WEKA package of
machine learning algorithms could be useful for data mining projects. Weka provides resources
for pre-analysis data cleaning, organization, regressing, grouping, connection discovery, and
visualization. Weka is freely available, open-source software that operates under the GNU Public
License. Additionally, it may be used to create novel methods for machine learning. The
algorithms may be called directly from Java code or executed on a dataset [18].

Table- 4: Theoretical results

Algorith Attribu Missi Prunin Outlier
m te Type ng g Detecti
Value Strate on
gy
ID3 Only No No Suscepti
categoric ble to
al outlier
values
CART Categori Yes Cost Can
cal and complexit handle
Numeric y pruning
al both is used
C4.5 Categori Yes Error Suscepti
cal and based ble to
Numeric pruning outlier
al both is used

In this study, three decision tree algorithms—ID3, C4.5, and CART—are tested on the same
datasets to see how they do. Table [17] below provides a concise summary of the three approaches'
results regarding runtime and accuracy. The splitting Criteria column details the algorithm's
division strategy for performance improvement. In the attribute type column, you can see what
sorts of data the algorithm can handle. The algorithm's performance may be evaluated by checking
whether the Missing Value field is filled in.

Table- 5: Practical results

Algorithm Time Taken Accuracy Precision
(Seconds) (%)
ID3 0.02 89.35 0.964
CART 0.5 97.11 0.972
C4.5 0.06 92.36 0.924

The following table shows the realistic output of three algorithms: ID3, C4.5, and CART. The
execution time for CART, ID3, and C4.5 is 0.5, 0.02, and 0.06 seconds, respectively. When
compared to ID3, CART has the slowest execution time. In spite of being the slowest method,
 Journal of Science and Technology
ISSN: 2456-5660 Volume 8, Issue 9 (Sep -2023)
www.jst.org.in DOI:https://doi.org/10.46243/jst.2024.v8.i9.pp1-8

CART delivers the most precise results, making it the best option. It seems that CART is the best
algorithm out of the three that were considered, according to the data presented in the table above.

Confusion Matrix:

Fig. 2 – Confusion matrix for ID3

Fig. 3 – Confusion matrix for C4.5

Fig. 4 – Confusion matrix for CART

III. CONCLUSION
Our decision tree methods of choice for this dataset were CART, ID3, and C4.5. For reliability,
speed, and precision, decision trees are the way to go. The recommendation system is heavily
relied upon by users to discover valuable material. After much discussion, the article's writers have
settled on the conclusion that, when tested on this dataset, CART achieves the highest levels of
accuracy and precision among decision tree methods.

REFERENCE

View publication stats

 Journal of Science and Technology
ISSN: 2456-5660 Volume 8, Issue 9 (Sep -2023)
www.jst.org.in DOI:https://doi.org/10.46243/jst.2024.v8.i9.pp1-8

Yes, for research papers [1]. You sadden me, my love. A survey of the research on multi-label
learning algorithms. Oregon State University in Corvallis, Oregon, on December 18, 2010.
[2]. Akcayol Yuldiz O., Utku A., and Uke Karacan Utku. Decision-Tree-Based Recommendation
System with Implicit Relevance Feedback Implementation. The citation is from the Journal of
Safety and Health, volume 10, issue 12, pages 1367–14. Last updated on December 1, 2015.
Contributed to by Gershman, Meisels, Lüke, Rokach, Schclar, and Sturm [3]. A Recommender
System Based on Decision Trees. Article published in Volume 17, Issue 3, by the International
Institute of Cognitive Science on June 3, 2010. I used "Jadhav SD" and "Channe HP" as references.
A collaborative filtering decision tree classifier is an effective recommendation system. Published
in 2016 by the International Journal of Engineering Research and Technology, volume 3, pages
2113–2118.
Everyone from Nürnberger and Genzmehr to Langer and Beel wrote a word or two here. We
provide Docear's approach of recommending scholarly articles in this article. Presented at the
ACM/IEEE-CS 2013 Digital Libraries Conference on July 22, 2013. The ACM, an organization
for modern languages. Zhou Xiang and Jiang S. Decision tree learning using a similarity split-off
criterion. As of 2012 Aug;7(8):1775-82, the paper was published in JSW. Mathuria, N., Bhargava,
N., Bhargava, R., and Sharma, G. (2013). An investigation into data mining decision trees using
the j48 algorithm. International Journal of Advanced Research in Computer Science and Software
Engineering, Volume 3, Issue 6, was released in June 2013 (IJARCSEE 2013).
several approaches to serial decision tree categorization and how to compare them. [8] Article by
Myanwu MN and Shiva SG published in the Journal of International Computer Science and
Security, volume 3, issue 3, pages 230–140, June 2009.

View publication stats