Academic Journal of Nawroz University (AJNU), Vol.11, No.

4, 2022
This is an open access article distributed under the Creative Commons Attribution License
Copyright ©2017. e-ISSN: 2520-789X
https://doi.org/10.25007/ajnu.v11n4a1636

An Approach based Iris Flower Species Recognition Using Machine

Learning Classifiers
Amar Yehya Hussien1
1
Department of Statistics, Van Yuzuncu Yil University

ABSTRACT
For data instances, classification is used to forecast group membership. Techniques for machine learning are being
introduced to make the classification problem simpler. The development of a model to categorize Iris blossoms using
an Artificial Neural Network (ANN) and Support Vector Machine (SVM) is the main goal of this paper. The Iris flower
data set is an example of a multivariate data set. It was first presented by Ronald Fisher, a British statistician and
biologist. Multi-parameter analysis of taxonomic is considered a main problem of iris flower recognition. The difficulty
is in differentiating between the iris flower species (Setosa, Versicolor, and Verginica) based on the dimensions of the
sepal and petal. The Iris data set would be classified by looking for patterns in the sepal and petal sizes of the Iris
flower, and then determining how the pattern was predicted to form the class of the Iris flower. Experimental results
illustrated that both ANN and SVM can classify iris flowers successfully by obtaining 98.66% and 97.79% of accuracy,
respectively.
Keywords: Machine Learning, Deep Learning, AI, Iris Flower.

1. Introduction Validated models are implemented in the last stage of

data mining approaches. Businesses may employ data
The world is surrounded by a variety of flora and
mining techniques to transform unusable data into facts.
animals in the modern world. People not understanding
Recognizing every aspect of consumer behavior will
their significance in our daily lives for numerous uses is
help businesses improve their communication strategies
the problem. There are lots of ayurvedic herbs, healing
and boost sales. Additionally, this information needs to
flowers, and nutritious foods in the area. This initiative
be appropriately categorized in order to maximize its
benefits many people by taking into account flowers as
value [5].
both decorations and remedies. Even the aroma is used
The goal of machine learning is to create computer
in numerous therapeutic applications in addition to
programs that can learn to improve and adapt as they
medicine [1]. The first section of this project is called
encounter new information. Predictive analytics or
Flower Classification, and it involves categorizing all the
statistical learning are other names for the research area
flowers according to their sepal, petal, and ovary
that lies at the nexus of statistics, Artificial Intelligence
dimensions. Due to its clarity and simplicity, the Iris
(AI), and computer science. Machine learning falls into
dataset is frequently used in machine learning and data
two major types. There are two types: supervised
science courses. Creating a method in which existing
learning and unsupervised learning. This particular
dataset-based data mining technologies are used to
paper focuses on supervised learning. The objective of
display all of the information about a flower's details [2].
supervised learning is to infer a function from labeled
Data mining combines a range of trends, clever methods,
training data. There are a number of training examples
algorithms, and software to sift user data and extract
in the training data. In supervised learning, each
information from enormous data warehouses. This
example consists of a pair of desired output values and
strategy will help businesses evaluate outcomes,
input objects. An inferred function is generated by a
anticipate possible trends, and forecast user behavior.
supervised learning algorithm from the training data,
Data mining uses four methods and phases for useful
which may then be used to map fresh samples.
data extraction. A database is a collection of data from
Regression and classification issues are within the
several sources, a sizable database that may contain issue
category of supervised learning challenges. When the
definitions. Data discovery is a technique for gleaning
output variable is a category, such as "illness" or "no
important information from enormous amounts of
disease," the situation is known as a classification
mysterious data [4]. Modeling, which is the third stage,
problem [6, 1].
comprises designing and assessing multiple templates.

469
Academic Journal of Nawroz University (AJNU), Vol.11, No.4, 2022

An approach for identifying Iris flower species is an original image of the flower.
described in this research. It operates in two stages: Patrick [10] concentrated on the dataset's statistical
testing and training. A Machine Learning (ML) method analysis using the iris flower example. They are
is loaded with the dataset of the training stage during examining two alternative approaches in his study. To
training, and labels are given to the data. The predictive identify the various classification patterns, the dataset is
model also indicates the species to which the iris flower plotted. Then, using a java program they developed,
belongs. Therefore, the predicted Iris species have been they may retrieve statistical data. In her research,
identified. This research focuses on machine learning- Poojitha [11] employed neural networks to examine data
based IRIS flower classification. The problem statement sets on iris flowers. A branch of computer science called
focuses on identifying IRIS flower species using machine learning. We have already loaded the iris
measurements of their floral attributes. The IRIS data set dataset and have divided it into three groups. They
would be classified by finding patterns in the petal and divided the dataset into groups using the k-means
sepal sizes of the IRIS flower and determining how the technique. Large-scale information aggregation is the
class of the IRIS flower was predicted by evaluating the main use of a neural network. Additionally, it is
pattern. In this study, we use data to train the machine employed in the mining of data, quantization of vectors,
learning model, and when previously unknown data is work approximation, division of images, and highlight
found, the algorithm predicts the species based on what extraction. Without any oversight, the findings are
it has learned from trained data. divided into three distinct iris species. Lakhdoura and
Elayachi [12] used WEKA 3.9 to do a test comparing the
2. Literature Review performance of two classifier methods: J48 (c4.5) and RF
The practice of categorizing distinct database objects into on the IRIS features. As a result, the University of
one or more groups or categories is known as data California, Irvine's ML library provides access to the IRIS
mining. The objective of the classification step is to plant dataset, one of the most popular datasets for
assign each instance to the relevant target class. This classification problems (UCI). The researchers also
section provides an overview of the most recent and contrasted the outcomes of both classifiers on numerous
practical classification methods that have been efficacy assessment metrics. According to the results, the
developed by researchers in the last two years across J48 classifier performs better than the Random Forest
many ML domains. Additionally, it only employs k- (RF) classifier for predicting IRIS variety using a range of
Nearest Neighbors, random forests, and decision trees as measures, including classification precision, mean
classifiers. absolute error, and construction time. The accuracy of
The Gaussian Naive Bayes technique is used by Zainab the J48 classifier is 95.83%, while that of the Random
Iqbal [7] to categorize the species of the iris flower. We Forest is 95.55%.
analyze the iris dataset using a scatter matrix and a Numerous research has been done using different
scatter plot that is constructed. The algorithm and methods to identify the species of the iris flower. Every
Python are both utilized in the paper to categorize the study employs a different method. The issue is the
many species of iris flowers. We can see that this categorization and identification of iris flower species
technique is effective for supervised learning based on their characteristics. After examining the
classification because it achieves a 95% accuracy rate. A characteristics of the iris flowers, we classify the iris data
C4.5 decision tree was suggested by Mijwil and Abttan set by identifying patterns, and we then project how the
[8] as a way to lessen the impacts of overfitting. IRIS, Car patterns will be processed to create the class of iris
Assessment, Bottle, and WINE were the datasets flowers. With the use of this classification and pattern,
utilized; both of these may be found in the UCI ML future predictions for unknown data can be made with
library. The issue with this classifier is that it overfits greater accuracy. The dataset for iris flowers is placed
because of its large number of nodes and divisions. It is into the machine learning prototype for the iris flower
possible that this overfitting will undermine the species approach.
classification system. The experimental results
demonstrated that, with an accuracy of roughly 92%, the 3. Developed Method
genetic algorithm was effective in reducing the effects of A data mining technique called classification divides
overfitting on the four datasets and increasing the data instances into a few different classes. The numerous
Confidence Factor (CF) of the C4.5 decision tree. Rong- models that built ML identification techniques have all
Guo Huang [9] focuses on flower detection using been developed to surpass one another. They all employ
Difference Image Entropy (DIE), a feature extraction- statistical techniques, including, but not limited to, DT,
based method. Their analysis of the experimental SVM, and ANN. These techniques look at the available
findings shows that the average recognition rate was data in various ways to guess [13]. Figure 1 displays the
95%. The DIE-based approach utilizes pre-processing conceptual framework that this work has built. The
and DIE computing to provide a recognition result from classification of iris flower species is done using a
470
Academic Journal of Nawroz University (AJNU), Vol.11, No.4, 2022

technique that is described. Testing and training are the order to categorize it. The classification of the iris flower
two stages. Machine learning models are fed data sets is then formed by looking closely at the pattern to make
during the training stage. Which category the iris flower the forecast. The machine learning model is trained by
belongs to is indicated by the predictive model. In this giving it data sets, and if any unexpected data is found
study, we use ML techniques to classify the many kinds or seen, it will forecast the species of flower based on
of iris flowers. By analyzing the sepal and petal sizes of what it has learned from training the data. Our work is
the flowers, we must find patterns in the iris data set in to identify iris species based on floral characteristics.

Data collection Normalization Data Splitting Training data

Classification Testing data Machine learning algorithm

Trained algorithm

Figure 1: Steps of the General Block Diagram

predictive approaches is to create a model that, in the
3.1 Normalization face of ambiguity, produces predictions based on data.
One of the most popular methods for preparing data is A predictive technique trains a system to produce an
normalization, which enables us to convert the values of accurate classification for the response to new data using
the dataset's numerical columns to a common scale. a known set of input data and a known response to the
Although it is not required for all available data for data (output). To remain in control over sensitive data
machine learning, normalization is used anytime a and to meet regulatory criteria, data classification is
dataset's properties have a wide range. A machine crucial. Understanding the data laws and rules that are
learning model's performance and dependability are used to categorize the data is crucial since machine
improved as a result. This study gave us the inspiration learning, a subset of artificial intelligence can assist in
to utilize the formula below, which allowed us to identifying and categorizing data [14].
transform the value into a suitable range.
A. Artificial Neural Network
𝑍𝑎𝑏 =
𝑍𝑎𝑏 − 𝑐𝑚𝑖𝑛
(1) A conceptual system or mathematical model based on
𝑐𝑚𝑎𝑥 − 𝑐𝑚𝑖𝑛
biological neural networks, or a replication of a
biological neural system, is known as an Artificial
Compute the cmin minimum value of the column. Neural Network (ANN). It uses a connectionist method
Subtract the cmin from each of the data, computer the of computation to handle information and is made up of
value of (cmax - cmin), and computer the new value by a network of artificial neurons. An ANN is often an
dividing the two previous results. adaptive system that modifies its structure in response
to information coming from the outside or inside the
Where cmin = minimum value of the column, cmax network [15] during the learning phase. Neural
=maximum value of the column, and Zab = Z (data item) networks are non-linear statistical data modeling tools,
to put it more simply. They can be applied to identify
present at athrow and bthcolumn. patterns in data or to model intricate relationships
between inputs and outputs. Similar to the extensive
3.2 Machine Learning Algorithm network of neurons in the human brain, a neural
ML is more efficient at validating the data and their network is a linked collection of nodes [16].
behavior and exploring information. When data is
accessible, divide it into training and test datasets, then
train an algorithm to investigate where the data will be
in the future. Machine learning employs two different
types of methods: descriptive techniques methods
(unsupervised methods), which uncover hidden
patterns or intrinsic structures in input data, and
predictive methods (supervised technique methods),
which train a framework on known input and output
data in order to predict future outputs. The goal of

471
Academic Journal of Nawroz University (AJNU), Vol.11, No.4, 2022

The simplest ANNs imitate the function of the brain by

using a layered structure of interconnected ANs. This
type of architecture is also referred to as a Dense or Fully
Connected Network since each neuron in an ANN
receives the activation of every AN of the preceding
layer as input parameters (Figure 2). A Feed-Forward
Network would be another term for this architecture,
which only connects layers in order. Deep ANNs are
ANNs having multiple layers between the input and
output layer (also known as hidden layers), and training
such networks is referred to as deep learning. Typically,
supervised learning is used for training, which involves
manually annotating the data before to training and
using backpropagation to change the weights for each Figure 3: Support Vector Machine [23]
input [17].
A task that can be divided into two categories is binary
categorization. SVM can be effectively employed when
the data contains exactly two categories. SVM begins by
categorizing data by locating hyperplanes and dividing
data points according to a few characteristics. There are
numerous hyperplanes; the best has the largest margins
and more interdependent data.

3.3 Training Method

Our method is trained using the dataset to make reliable
Figure 2: Basic Structure of ANN [16] output predictions. By extracting information from this
dataset, we concentrate on categorizing the iris flower
B. Support Vector Machine class. The processed data includes an analysis of each
The main application of SVM in medical data processing parameter. Data preparation is necessary for the
is identification. In a multidimensional setting, SVM machine learning process in order to transform the data
functions as a separator of distinct data sets. It has the into a format that the machine can interpret. We may say
ability to classify data into binary and multiclass that the algorithm can now quickly analyze the data
categories. Clinical coding, which converts medical data features. The data is transformed into binary format for
into standardized statistical code, depends on our project. The algorithm now performs the necessary
identification. To assess crucial data, identification, for calculations. In order for us to understand the output, it
instance, splits the data into diagnostic or process codes. is converted from binary to hexadecimal format.
The identification is based on different health Hexadecimal codes are used to identify colors. The aim
assessment-related characteristics [18]. The main is to categorize the flowers according to their flower
objective of this strategy, as shown in Figure 3, is to characteristics.
project nonlinearly identifiable samples to a higher-
dimensional space using various kernel functions. 4. Evaluation Results
Recently, the popularity of SVM has drawn a lot of The iris dataset that we gathered is made up of 150 tuples
attention to kernel techniques. The connection between with four flower parameters, including sepal length,
linearity and nonlinearity mainly relies on kernel sepal width, petal length, and petal width. Setosa,
functions. SVM employs the idea of transforming the Versicolor, and Virginia are the 3 distinct flower species
input domain into a space with many dimensions to that are represented. The dataset has been divided into
enhance the identification function [19]. training and testing, 75% of the dataset is the training
dataset, while 25% is the testing dataset.

472
Academic Journal of Nawroz University (AJNU), Vol.11, No.4, 2022

4.1 Dataset The observational data with 150 items are included in the
The UCI Machine Learning Repository served as the Iris flower data collection. Since the data frame contains
source of the dataset for this study. A multivariate data 150 items that fall into one of the three target categories
set is the Iris flower data set, often known as Fisher's Iris and four features (sepal width, sepal length, petal width,
data set. The data set includes 50 samples from each of and petal length). In this stage, we analyze the dataset
the three Iris species (Iris virginica, Iris versicolor, and mathematically to evaluate the performance of the
Iris setosa). Each sample was measured (in centimeters) method. Examples from the IRIS dataset is shown in
for the following four characteristics: sepal length, sepal Table 1 as illustrations.
width, sepal length, and sepal length. The Iris species are
depicted in Figure 4.

Figure 4: Types of Irises Flowers [11]

Table 1: IRIS Dataset-based Sample Instances

Sepal length Sepal width Petal length Petal width Species

5.1 3.5 1.4 0.2 setosa
4.9 3.0 1.4 0.2 setosa
4.7 3.2 1.3 0.2 setosa
7.0 3.2 4.7 1.4 versicolor
6.4 3.2 4.5 1.5 versicolor
6.9 3.1 4.9 1.5 versicolor
6.5 3.0 5.2 2.0 virginica
6.2 3.4 5.4 2.3 virginica
5.9 3.0 5.1 1.8 virginica

4.2 Evaluation Results

The input data is the 150x4 matrix from the IRIS dataset,
which may be acquired from the UCI repository at
www.ics.uci.edu. 75 of these 150 instances were utilized
for testing, while 75 were utilized for training. The IRIS
dataset's columns and rows each indicate a characteristic
or property of the iris flower. Sepal length, sepal width,
petal length, and petal breadth are the four
characteristics of an iris flower. Each sample in
supervised learning has a class label, therefore the Iris
dataset contains three classes: Setosa, Versicolor, and
Virginia. On the Iris dataset, we can learn the following
interpretable rule. Figure 5: Representing IRIS Dataset

473
Academic Journal of Nawroz University (AJNU), Vol.11, No.4, 2022

The preceding rule can be parsed in the manner listed boost their efficiency and achieved 95.83% accuracy for
below. If an iris flower complies with the criterion, it is j48 and 95.55% accuracy for RF. The study in [21]
assumed to be an Iris Versicolor. The rule is a disjunction presents a method for iris classification using different
of literals in each of the three clauses that make up the machine learning techniques including LDA, PCA, RF,
conjunction. A literal can make binary decisions if the and LR. The same dataset has been used in the
sepal length is greater than 6.3. According to the evaluation; results showed that LDA among them
aforementioned rule, a clause is satisfied when at least achieved higher accuracy by obtaining 100% whereas
one literal is met, and the rule is met when every clause other techniques achieved lower accuracy compared to
is met. In this work, we are interested in learning logical this work by obtaining 86%, 94%, and 96% respectively.
formulations that represent categorization rules. Moreover, a study in [22] used only DT for iris
Equations (2–4) were used to calculate accuracy, as classification by obtaining 98%. This method obtained
shown below. higher accuracy compared to this study when using
𝐴𝑐𝑐𝑢𝑟𝑎𝑐𝑦 =
𝑇𝑃+𝑇𝑁
(2) SVM while when using ANN this work obtained higher
𝑇𝑃+𝑇𝑁+𝐹𝑃+𝐹𝑁
𝑇𝑃 accuracy.
𝑅𝑒𝑐𝑎𝑙𝑙 = (3)
𝑇𝑃+𝐹𝑁
𝑇𝑃
𝑃𝑟𝑒 = × 100% (4) 5. Conclusion
(𝑇𝑃+𝐹𝑃)
The terminology used in the above equations that were In this research, we trained our data using a variety of
listed in the confusion matrix is clarified in more detail strong algorithms. The best results can only be obtained
below: True Positive (TP) is the result of the sample by data processing, and as we can see from the findings
properly predicting the positive class; False Positive (FP) above, those results are quite pleasing. In two of the four
is an outcome in which the sample correctly predicts the models mentioned above, which have accuracy scores of
positive class, whereas False Negative (FN) is an event 98.66% and better respectively, it is possible to predict
in which the sample wrongly predicts the negative class. the species of the iris flower. We can draw the conclusion
True Negative (TN) is an outcome in which the sample that it will be feasible to identify the species of any flower
accurately predicts the negative class. in the future with the right information on its
characteristics.
Table 2: Evaluation Results on IRIS Classification Based
data Splitting with 75% Training and 25% Testing References
Classifier Accuracy Recall Precision [1] Bhutada, S., K. Tejaswi and S. Vineela. Flower Recognition Using
Machine Learning. International Journal Of Researches In
ANN 98.66 96.32 98
Biosciences, Agriculture And Technology, 4(2), 67-73, 2021.
SVM 96.79 93.31 93.88 [2] Khalid, L. F., Abdulazeez, A. M., Zeebaree, D. Q., Ahmed, F. Y.,
ANN and SVM, are two classifiers that were assessed & Zebari, D. A. (2021, July). Customer churn prediction in
using this method using IRIS datasets. The findings telecommunications industry based on data mining. In 2021 IEEE
Symposium on Industrial Electronics & Applications (ISIEA) (pp. 1-
show that the classifiers offer various resolutions on
6). IEEE.
various datasets due to functional differences. The [3] Haji, S. H., Abdulazeez, A. M., Zeebaree, D. Q., Ahmed, F. Y., &
evaluation of the accuracy, precision, and recall-based Zebari, D. A. (2021, July). The Impact of Different Data Mining
method was described in Table 2. The ANN algorithm Classification Techniques in Different Datasets. In 2021 IEEE
outperforms both ANN and SVM in terms of Symposium on Industrial Electronics & Applications (ISIEA) (pp. 1-
performance. This study's final finding is that ANN 6). IEEE.
outperforms SVM in terms of accuracy. [4] P. Galdi and R. Tagliaferri, “Data mining: accuracy and error
measures for classification and prediction,” Encycl. Bioinforma.
Comput. Biol., pp. 431–6, 2018.
Table 3: Evaluation Results on IRIS Classification Based [5] Chicho, B. T., Abdulazeez, A. M., Zeebaree, D. Q., & Zebari, D.
data Splitting with 80% Training and 20% Testing A. (2021). Machine learning classifiers-based classification for IRIS
Classifier Accuracy Recall Precision recognition. Qubahan Academic Journal, 1(2), 106-118.
ANN 96.72 95.09 96.72 [6] Rao, T. S., Hema, M., Priya, K. S., Krishna, K. V., & Ali, M. S.
(2021). Iris Flower Classification Using Machine
SVM 95.03 93.01 92.31 Learning. Network, 9(6).
Based on the classifier employed in this study, ANN and [7] Shilpi Jain, V Poojitha, “By Using Neural Network Clustering
SVM classification algorithms were documented in tool in MATLAB Collecting the IRIS Flower”, Proc. IEEE , vol. 109,
linked works in this article, illuminating the significant 2020.
tasks that the researchers set with each approach [8] M. M. Mijwil and R. A. Abttan, “Utilizing the Genetic
Algorithm to Pruning the C4. 5 Decision Tree Algorithm,” Asian J.
examined. In this section, the findings from this study
Appl. Sci. ISSN 2321– 0893, vol. 9, no. 1, 2021.
were contrasted with those from studies based on [9] Roung– Guo Huang, Sang-Hyeon Jin, Jung –Hyun Kim and
research from related works. According to related work, Kwang- Seck Hong, “Flower Image Recognition Using Difference
a study [20] employed j48 and RF on IRIS datasets to Image Entropy”. DOI: 10.1145/1821748.1821868

474
Academic Journal of Nawroz University (AJNU), Vol.11, No.4, 2022

[10] K R Rathy, Arya Vaishali, “Classification of Dataset using

Efficient Neural Fuzzy Approach”, vol. 099, August 2019.
[11] D. Decoste, E. Mjolsness. 2001. “State of the art and future
prospects by using Machine Learning”, vol. 320, 2013.
[12] Y. Lakhdoura and R. Elayachi, “Comparative Analysis of
Random Forest and J48 Classifiers for ‘IRIS’ Variety Prediction,”
Glob. J. Comput. Sci. Technol., 2020
[13] L. Dhanabal and S. P. Shantharajah, “A study on NSL-KDD
dataset for intrusion detection system based on classification
algorithms,” Int. J. Adv. Res. Comput. Commun. Eng., vol. 4, no. 6,
pp. 446–452, 2015.
[14] Hassan, C. A. U., Khan, M. S., & Shah, M. A. (2018,
September). Comparison of machine learning
algorithms in data classification. In 2018 24th
International Conference on Automation and Computing
(ICAC) (pp. 1-6). IEEE.
[15] Taher, K. I., Abdulazeez, A. M., & Zebari, D. A.
(2021). Data Mining Classification Algorithms for
Analyzing Soil Data. Asian Journal of Research in Computer
Science, 17-28.
[16] Zafeiris, D.; Rutella, S.; Ball, G.R. An Artificial
Neural Network Integrated Pipeline for Biomarker
Discovery Using Alzheimer’s Disease as a Case Study.
Comput. Struct. Biotechnol. J. 2018, 16, 77–87.
[17] Zebari, D. A., Abrahim, A. R., Ibrahim, D. A.,
Othman, G. M., & Ahmed, F. Y. (2021). Analysis of Dense
Descriptors in 3D Face Recognition. In 2021 IEEE 11th
International Conference on System Engineering and
Technology (ICSET) (pp. 171-176). IEEE.
[18] Abdulqadir, H. R., Abdulazeez, A. M., & Zebari, D.
A. (2021). Data mining classification techniques for
diabetes prediction. Qubahan Academic Journal, 1(2),
125-133.
[19] Ibrahim, D. A., Zebari, D. A., Ahmed, F. Y., &
Zeebaree, D. Q. (2021, November). Facial Expression
Recognition Using Aggregated Handcrafted Descriptors
based Appearance Method. In 2021 IEEE 11th
International Conference on System Engineering and
Technology (ICSET) (pp. 177-182). IEEE.
[20] Y. Lakhdoura and R. Elayachi, “Comparative
Analysis of Random Forest and J48 Classifiers for ‘IRIS’
Variety Prediction,” Glob. J. Comput. Sci. Technol., 2020.
[21] D. Rana, S. P. Jena, and S. K. Pradhan, “Performance
Comparison of PCA and LDA with Linear Regression
and Random Forest for IRIS Flower Classification,”
PalArchs J. Archaeol. EgyptEgyptology, vol. 17, no. 9, pp.
2353–2360, 2020.
[22] K. Sarpatwar et al., “Privacy Enhanced Decision Tree
Inference,” in Proceedings of the IEEE/CVF Conference on
Computer Vision and Pattern Recognition Workshops, 2020,
pp. 34–35.
[23] Abdulazeez, A. M., Zeebaree, D. Q., Zebari, D. A., &
Hameed, T. H. (2021). Leaf Identification Based on
Shape, Color, Texture and Vines Using Probabilistic
Neural Network. Computación y Sistemas, 25(3), 617-631.

475