IAES International Journal of Artificial Intelligence (IJ-AI)

Vol. 11, No. 3, September 2022, pp. 1164~1174

ISSN: 2252-8938, DOI: 10.11591/ijai.v11.i3.pp1164-1174  1164

A proposed model for diabetes mellitus classification using

coyote optimization algorithm and least squares support vector
machine

Baydaa Sulaiman Bahnam1, Suhair Abd Dawwod2

1
Department of Software, College of Computer Science and Mathematics, Mosul University, Mosul, Iraq
2
Department Management Information Systems, College of Administration and Economics, Mosul University, Mosul, Iraq

Article Info ABSTRACT

Article history: One of the most dangerous health diseases affecting the world's population is
diabetes mellitus (DM), and its diagnosis is the key to its treatment. Several
Received Oct 27, 2021 methods have been implemented to diagnose diabetes patients. In this work,
Revised May 19, 2022 a hybrid model which combines of coyote optimization algorithm (COA)
Accepted Jun 17, 2022 and least squares support vector machine (LS-SVM) is proposed to classify
of Type-II-DM patients. LS-SVM classifier is applied for classification
process but it's very sensitive when its parameter values are changed. To
Keywords: overcome this problem, COA algorithm is implemented to optimize
parameters of the LS-SVM classifier. This is the goal of the proposed model
Classification called the COA-LS-SVM. The proposed model is implemented and
Coyote optimization algorithm evaluated using Pima Indians Diabetes Dataset (PIDD). Also, it's compared
Diabetes mellitus with several classification algorithms that were implemented on the same
Least squares support vector PIDD. The experimental results demonstrated the effectiveness of the
machine proposed model and its superiority over other algorithms, as it could
Swarm intelligence algorithms accomplish an average classification accuracy of 98.811%.
This is an open access article under the CC BY-SA license.

Corresponding Author:
Baydaa Sulaiman Bahnam
Department of Software, College of Computer Science and Mathematics, Mosul University
Mosul, Iraq
Email: [email protected]

1. INTRODUCTION
Diabetes mellitus (DM) is one of the most diffuse diseases in the world. At present, about
425 million people have been infected worldwide, and it is expected that up to 700 million people will be
infected by 2045 [1]. It is a chronic metabolic disease caused by the pancreas not producing enough insulin
or the body's cells not responding to the insulin that is produced. Thus, high blood sugar occurs, which leads
to many health disorders. Depending on World Health Organization (WHO) and American Diabetes
Association (ADA), DM is classified into four types [2]–[4]: i) type-I-DM: or insulin-dependent diabetes
mellitus (IDDM). The failure of the body to produce insulin due to the destruction of the pancreas generates
this type. It is usually diagnosed in children and young age, approximately 5%-10% of all diabetes mellitus
are of this type; ii) type-II-DM: also called non-insulin-dependent diabetes mellitus (NIDDM) or "adult-onset
diabetes." It is the most common, about 90% of diabetics. It results from the failure of the body's cells from
consuming the secreted insulin and thus leads to an increase in blood sugar levels; iii) gestational diabetes
(GDM): 4% of pregnant women develop this type due to pregnancy changes in the body, and it usually turns
into the second type after pregnancy; iv) rare specific diabetes. it is caused by genetic and metabolic
disorders.

Journal homepage: http://ijai.iaescore.com

Int J Artif Intell ISSN: 2252-8938  1165

Diabetes is one of the main reasons for the increase in the number of deaths in the world, especially the
Type-II-DM, which is the most common [4]. Many serious health disorders occur when neglected and not
treated, such as heart attack, myocardial infarction, stroke, renal failure, blindness, neuropathy, gangrene, micro
vascular damage and increased susceptibility to infection [5], [6]. With its spread, places a great strain on the
public health system [7], [8]. So, the important step is to detect and diagnose it early. In the modern times, a lot
of research works concentrated on using machine learning (ML) algorithms to detect and diagnose of DM using
pima Indian diabetes dataset (PIDD) [9]–[11]. A study of Patel and Tamani [11] showed that the accuracy of the
logistic regression (LR) and gradient boost (GB) algorithms were higher than the other algorithms at 79%.
Patil et al. [12] proposed approach depending on Mayfly algorithm for feature selection and support vector
machine (SVM) classifier to diagnose Type-II-DM. The outcome showed that the accuracy of this approach is
94.5% comparing with other studies. Panda et al. [13] used four algorithms of ML: SVM, k-nearest neighbor
(KNN), LR, and gradient boost (GB) to predict DM. The results showed that the GB algorithm outperforms
the other algorithms with the highest accuracy of 81.25%. Alalwan [14] proposed two conceptual models of
data mining: self organizing map (SOM) and random forest algorithm (RFA). The experimental showed that the
accuracy of SOM is outperformed RFA, which reached 85%. Rajni and Amandeep [15] used the RB-Bayes
algorithm, which reached 72.9% the highest prediction accuracy compared to other algorithms. Bozkurt et al.
[16] used six various neural networks to classify DM patients. The experiments showed that distributed time
delay net-works (DTDN) is the best comparing with others with accuracy of 76.00%. Rahman and Afroz [17]
used data mining tools to comparative study of different classification techniques. These techniques are
multilayer perceptron (MLP), bayesnet, naïve bayes, J48graft, fuzzy lattice reasoning (FLR), JRip (JRipper),
fuzzy inference system (FIS) and adaptive neuro-fuzzy inference system (ANFIS). The results showed that
J48graft classifier is best with an accuracy of 81.33%. Khashei et al. [18] constructed a hybrid model of MLP
which depended on the idea of soft computing and artificial intelligence techniques. The experiments showed
that the hybird model MLP is outperform over the other methods with accuracy 81.2%. Marcano-Cedeño et al.
[19] proposed a prediction model AM-MLP that based on artificial metaplasticity (AM) with MLP to predicate
diabetes. The accuracy obtained from this model was 89.93%. Karegowda et al. [20] presented a hybrid
approach GA-BPN that combines genetic algorithm (GA) and back propatation network (BPN). The GA was
used to optimize the weight of BPN. The accuracy of the GA-BPN model was 84.713% which was better than
without GA. Fiuzy et al. [21] proposed a model based on three techniques which are: fuzzy system to instant
and precise decision making, ant colony algorithm (ACO) to select best rules in fuzzy systems while ANN for
modeling, structure identification and parameter identification. The accuracy reached from this model was
95.852%. Haritha et al. [22] used firefly and cuckoo search algorithms to reduce dimension and then classify
UCI dataset type I and type II using traditional KNN classifier and fuzzy KNN. The accuracy obtained of UCI
type II is 71.3% for firefly-fuzzy-KNN and 74.8% for cuckoo-fuzzy-KNN. Zhang et al. [23] used a multi-layer
feed-forward neural network to predict of DM. This network provided results with 82% accuracy.
The main objective of this paper is to propose a hybrid model for diagnosing diabetes to increase
health awareness in the community with the help of health practitioners in diagnosing the disease to control it
and avoid its danger. This proposed model coyote optimization algorithm and least squares support vector
machine (COA-LS-SVM) is based on the COA algorithm and the LS-SVM classifier. Where the COA
algorithm uses to find the optimal values for the LS-SVM parameters to overcome its sensitivity to changes in
its parameter values and the LS-SVM classifier uses to classify Type-II-DM. Achieving a balance between the
exploration and exploitation distinguishing the algorithm of COA from others during the optimization process.
For this reason, the authors in this paper motivated to use this algorithm for the first time to find the optimal
values of LS-SVM parameters to overcome the problem of its sensitivity to changes in its parameter values.
Also, this study compares the performance of this proposed approach with others ones. The implementation
results demonstration the powerful of this proposed model COA-LS-SVM, which has the average accuracy of
98.811% outperforms the other algorithms. The rest of the paper is structured; next section presents COA and
LS-SVM algorithms. Section 3 described the proposed model and the data set. The experimental results are
covered in section 4. Finally, the conclusion and future work of this paper are mentioned in section 5.

2. OVERVIEW OF METHODOLOGIES
The present work intends to propose a model a model COA-LS-SVM based on COA algorithm and
LS-SVM classifier. The proposed model was used to classify DM patient accurately. The next two sections
provide an overview of these algorithms used.

2.1. Coyote optimization algorithm (COA)

The swarm intelligence algorithms (SIAs) are inspired from the social action of creatures to solve
several problems [24]–[26]. One of these recent algorithm is COA for global optimization problems. It was
lately proposed meta-heuristic algorithm by Pierezan et al. in 2018 [27]. The major scheme of COA
A proposed model for diabetes mellitus classification … (Baydaa Sulaiman Bahnam)
1166  ISSN: 2252-8938

optimizer is stimulated on Canis latrans species that stay principally in North America [27]. This algorithm is
concerned to represent the coyotes’ social society and acclimatize it with a various algorithmic structure. An
important advantage of this method is that it maintains a balance between the exploitation and exploration
phase through this optimization approached [28]. COA is not attentive with the hierarchy and dominance
rules pursued in grey wolf optimization (GWO) animals, and also it does not rely only on the hunting prey
pursued in the GWO, but on the social structure and exchange of methodical experiences among wolves. By
moving towards the prey as a group, it made it characterized by a cooperative trait while devouring the prey
individually [29]. And coyotes can locate prey through their strong sense of smell. The hunt takes place as
the coyotes attack the prey in groups, and this requires the agents to update their positions to the best. When
Coyotes’ infecting their rivals, they are fully ready with a threat chance and flit to new position as excessive
random distance away from its current position. Consider the following given to start with COA algorithm
[30]–[33]:
𝑝,𝑡
The COA technique has been prepared based on the social conditions 𝑠𝑜𝑐𝑐 of 𝑐 𝑡ℎ coyote in 𝑝𝑡ℎ
pack at 𝑡 𝑡ℎ instant of time for the decision variables 𝑥⃗ which can be written:
𝑝,𝑡
𝑠𝑜𝑐𝑐 = 𝑥⃗ = (𝑥1 , 𝑥2 , … , 𝑥𝐷 ) (1)

where, 𝐷 is the dimension of the search space. The COA starts with setting coyotes’ global population, the
𝑝,𝑡
social condition, is 𝑠𝑜𝑐𝑐,𝑗 for the 𝑗𝑡ℎ dimension which can be written:

𝑝,𝑡
𝑠𝑜𝑐𝑐,𝑗 = 𝑙𝑏𝑗 + 𝑟𝑗 (𝑢𝑏𝑗 − 𝑙𝑏𝑗 ) (2)

where, 𝑟𝑗 ∈ [0,1] is the real random number, 𝑙𝑏𝑗 𝑎𝑛𝑑 𝑢𝑏𝑗 are the lower and upper bounds of the 𝑗𝑡ℎ decision
variable. The fitness function of each coyote in their current social conditions is calculated in (3):
𝑝,𝑡 𝑝,𝑡 𝑝,𝑡
𝑓𝑖𝑡𝑐 = 𝑓𝑐 (𝑠𝑜𝑐𝑐 ) (3)

Randomly, the algorithm updates the packs location. As well as the candidates update their position by
departing their packs to other one. This behavior can be represented by the following Probability 𝑃𝑒 which
based on 𝑁𝑐 :

𝑃𝑒 = 0.005. 𝑁𝑐2 (4)

where the number of 𝑁𝑐 that makes 𝑃𝑒 > 1 is restricted to 14 coyotesinner the pack. The alpha coyote as in
(5) which is the best solution of each iteration. It means that the alpha coyote is only one for the global
population to optimize the problem in 𝑝𝑡ℎ pack at 𝑡 𝑡ℎ instant of time:
𝑝,𝑡 𝑝,𝑡
𝑎𝑙𝑝ℎ𝑑 𝑝,𝑡 = {𝑠𝑜𝑐𝑐 |𝑎𝑟𝑔𝑐={1,2,…𝑁𝑐} 𝑚𝑖𝑛𝑓(𝑠𝑜𝑐𝑐 )} (5)

All the coyotes’ information in COA are linked and calculated as culture transformation as the following:
𝑝,𝑡
𝑂(𝑁𝑐+1)⁄2,𝑗 , 𝑁𝑐 𝑖𝑠 𝑜𝑑𝑑
𝑝,𝑡 𝑝,𝑡 𝑝,𝑡
𝑐𝑢𝑙𝑡𝑗 ={ 𝑂𝑁 +𝑂 𝑁 (6)
𝑐/2,𝑗 ( 2𝑐 +1),𝑗
, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
2

where 𝑂𝑝,𝑡 , is the ordered social conditions of coyotes 𝑝𝑡ℎ pack at 𝑡 𝑡ℎ instant of time. The birth and death of
𝑝,𝑡
a coyote in COA are two important happenings, as this coyote' age is the 𝑎𝑔𝑒𝑐 ∈ 𝑁. The birth of a new
coyote is affected by the social conditions surrounding the randomly chosen parents, as well as the influence
of the environment, such as (7):
𝑝,𝑡
𝑠𝑜𝑐𝑟1,𝑗 , 𝑟𝑛𝑑𝑗 < 𝑃𝑠 𝑜𝑟 𝑗 = 𝑗1
𝑝,𝑡 𝑝,𝑡
𝑝𝑢𝑝𝑗 = { 𝑠𝑜𝑐𝑟2,𝑗 , 𝑟𝑛𝑑𝑗 ≥ 𝑃𝑠 + 𝑃𝑎 𝑜𝑟 𝑗 = 𝑗2 (7)
𝑅𝑗 , 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒

Int J Artif Intell, Vol. 11, No. 3, September 2022: 1164-1174

Int J Artif Intell ISSN: 2252-8938  1167

where 𝑟1 𝑎𝑛𝑑 𝑟2 are random coyotes from 𝑃𝑡ℎ pack, 𝑗1 𝑎𝑛𝑑 𝑗2 are random dimensions of the problem,
𝑃𝑠 𝑎𝑛𝑑 𝑃𝑎 are scatter and association probabilities respectively that declare the coyote’s cultural diversity
from the pack, 𝑅𝑗 is random number within the bounds of the 𝑗𝑡ℎ decision variable and 𝑟𝑛𝑑𝑗 is random
number in [0,1] generated by uniform probability. The cultural diversity of the coyotes in the pack described
by 𝑃𝑠 𝑎𝑛𝑑 𝑃𝑎 , which can be calculated:

𝑃𝑠 = 1⁄𝐷 (8)

𝑃𝑎 = (1 − 𝑃𝑠 )⁄𝐷 (9)

There are three rules for life cycle of COA as shown in the pseudo-code-1 [31]–[36]:

The pseudo-code-1. Life cycle rules of COA

Calculate 𝜔 and 𝜑
(𝜔 is the worst fitness function of the coyotes; 𝜑 is the coyotes’ number in pack)
If 𝜑 = 1
Parent survive while the only coyote in 𝜔 dies
Else if 𝜑 > 1
Parent survive while the oldest coyote in 𝜔 dies
Else
Parent die
End if

The cultural adaptation among the packs is determined by two factors alpha influence 𝛿1 and pack
influence 𝛿2 as:
𝑝,𝑡
𝛿1 = 𝛼 𝑝,𝑡 − 𝑠𝑜𝑐𝑐𝑟1 (10)

𝑝,𝑡
𝛿2 = 𝑐𝑢𝑙 𝑝,𝑡 − 𝑠𝑜𝑐𝑐𝑟2 (11)

where, 𝑐𝑟1 𝑎𝑛𝑑 𝑐𝑟2 are the random coyotes. To update the social condition of the coyote is wrriten as:
𝑝,𝑡 𝑝,𝑡
𝑛𝑒𝑤 − 𝑠𝑜𝑐𝑐 = 𝑠𝑜𝑐𝑐 + 𝑟𝑑1. 𝛿1 + 𝑟𝑑2. 𝛿2 (12)

where, 𝑟𝑑1 𝑎𝑛𝑑 𝑟𝑑2 are random numbers in the range [0,1]. Finally, the new fitness function and the
updating process of the social condition are founded by (13) and (14) respectively:
𝑝,𝑡 𝑝,𝑡
𝑛𝑒𝑤 − 𝑓𝑖𝑡𝑐 = 𝑓(𝑛𝑒𝑤 − 𝑠𝑜𝑐𝑐 ) (13)

𝑝,𝑡 𝑝,𝑡 𝑝,𝑡

𝑝,𝑡+1 𝑛𝑒𝑤 − 𝑠𝑜𝑐𝑐 , 𝑛𝑒𝑤 − 𝑓𝑖𝑡𝑐 < 𝑓𝑖𝑡𝑐
𝑠𝑜𝑐𝑐 = { 𝑝,𝑡 (14)
𝑠𝑜𝑐𝑐 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒

The following is the pseudo-code-2 that illustrates COA [34]-[36]:

The pseudo code-2. Coyote optimization algorithm COA

Determine population Np and coyote Nc by (2)
Find the Fitness function of the coyote by (3)
While stop criteria is not meet do
For each population P do
Determine the alpha coyote by (5)
Calculate the culture transformation by (6)
For coyote C of each population P do
Find the new social condition by (12)
Find the new fitness function by (13)
Update the social condition by (14)
End for
Perform the birth and death process by (7) and pseudo-code-1
End for
Perform pack's transitions by (4)
Update the age of coyotes
End while
Output the global best coyote
2.2. Least squares support vector machine (LS-SVM)

A proposed model for diabetes mellitus classification … (Baydaa Sulaiman Bahnam)

1168  ISSN: 2252-8938

One of the versions for SVM classifiers is LS-SVM classifier, which suggested by Suykens
Vandewalle in 1999 [37]. The goal of LS-SVM classifier is to detect optimal separating hyper-plane in
higher dimensional space by using euclidean distance [37], [38]. The advantage of LS-SVM is that it can
handle a set of linear equations instead of the quadratic programming problem that suffers from high
arithmetic operations [39]. It is famous for its extreme sensitivity to a change in the values of its parameters.
Consider the following given to start with LS-SVM [39]–[41]:
In the primal weight space, the optimization problem is formulated in (15), if we
consider {𝑥𝑘 , 𝑦𝑘 }𝑁 𝑛
𝑘=1 is a training set of N points, in which 𝑥𝑘 ∈ 𝑅 for input data and 𝑦𝑘 ∈ 𝑅 for output data:

1 1
𝑀𝑖𝑛 𝐽(𝑤, 𝑏)𝑤,𝑏,𝑒 = 𝑤 𝑇 𝑤 + 𝛾 ∑𝑁 2
𝑘=1 𝑒𝑘 (15)
2 2

yield to:
𝑦𝑘− (𝑤 𝑇 𝜑𝑥𝑘 + 𝑏) = 𝑒𝑘 , 𝑘 = 1,2, … . , 𝑁 (16)

where 𝛾 is regularization factor, 𝑒𝑘 is the difference between the desired output 𝑦𝑘 and the actual output,
𝜑(. ) is nonlinear function, 𝑤 is weight vector and 𝑏 is bias term, where 𝑏 ∈ 𝑅. Also, a linear classifier in
the new area takes as in (17):

𝑦(𝑥) = 𝑠𝑖𝑔𝑛(𝑤. 𝜑(𝑥) + 𝑏) (17)

Calculating the duple area instead of the initial area by finding the following Lagrangian function:

𝐿(𝑤, 𝑒, 𝛼) = 𝐽(𝑤, 𝑒) − ∑𝑁 𝑇
𝑘=1 ∝𝑘 (𝑤 𝜑(𝑥𝑘 ) + 𝑒𝑘 − 𝑦𝑘 ) (18)

where ∝𝑘 is Lagrangian multipliers called support vectors. An objective function in (18) is optimal when it
satisfies the following conditions of karush-kuhn-tucker (KKT) in (19):

𝛿𝐿
= 0 → 𝑤 = ∑𝑁
𝑘=1 𝛼𝑘 𝑦𝑘 𝜑(𝑥𝑘 ) (19)
𝛿𝑤

𝛿𝐿
= 0 →∝𝑘 = 𝛾𝑒𝑘 , 𝑘 = 1, … . . , 𝑁
𝛿𝑤
𝛿𝐿
= 0 → 𝑤 𝑇 𝜑(𝑥𝑘 ) + 𝑒𝑘 − 𝑦𝑘 = 0, 𝑘 = 1, … . , 𝑁
𝛿𝑤

The following linear system could obtain after removal of 𝑤 and 𝑒:

1
(𝐾 + ) ∝= 𝑦
𝜎
𝑝,𝑡 𝑝,𝑡 𝑝,𝑡
𝑝,𝑡+1 𝑛𝑒𝑤 − 𝑠𝑜𝑐𝑐 , 𝑛𝑒𝑤 − 𝑓𝑖𝑡𝑐 < 𝑓𝑖𝑡𝑐
𝑠𝑜𝑐𝑐 = { 𝑝,𝑡 (20)
𝑠𝑜𝑐𝑐 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒

where the Kernel Matrix is 𝑦 = [𝑦1 , 𝑦2 , … , 𝑦𝑘 ]𝑇 , 𝛼 = [𝛼, 𝛼2 , … , 𝛼𝑘 ]𝑇 where 𝐾 ∈ 𝑅𝑁𝑥𝑁 , 𝜎 is Gaussian Kernel
function. The function estimation has been obtained as the result of LS-SVM model in (21):

𝑦(𝑥) = ∑𝑁
𝑘=1 𝛼𝑘 𝐾(𝑥, 𝑥𝑘 ) (21)

and, to perform LS-SVM, the radial basis function (RBF) has been utilized:

|𝑥−𝑥𝑘 |2
𝐾(𝑥, 𝑥𝑘 ) = exp (− ) (22)
𝜎2

The following is the pseudo-code-3 that illustrates LS-SVM [42]–[45]:

The pseudo code-3. LS-SVM Algorithm

Enter the data set of n data point {xk , yk }N
k=1 , where xi is the i
th input vector and yi ∈ R is
the corresponding ith target with values {-1,+1}.
For each enter data point, randomly generate weights.

Int J Artif Intell, Vol. 11, No. 3, September 2022: 1164-1174

Int J Artif Intell ISSN: 2252-8938  1169

For each enter data point, randomly set the initial bias b and error e.
Randomly set the initial values of γ and σ.
Calculate and look for the values of (w, b, e) that minimize the objective function using
(15) and (16).
Calculate the function of Lagrangian using (18) with the solution, which must meet the
conditions of KKT in a group of (19).
Calculate the number of support vectors ∝ using (20).
Classify the training data of LS-SVM using (21) with Kernel function RBF using (22).
Classify any new data point using (17) and Kernel function RBF using (22).
Repeat till stopping criteria is met, usually till reach the maximum number of iterations.

3. METHODOLOGY
3.1. The proposed algorithm COA-LS-SVM
The proposed algorithm is a combination of two algorithms COA and LS-SVM as shown in the
Figure 1. Where COA algorithm was used in the first stage in order to obtain the optimal parameters for LS-
SVM, while in the second stage LS-SVM classifier was used to classify patients as: i) first stage: optimizing
parameters. The goal of this stage is to obtain the optimal parameter values of LS-SVM. The COA algorithm
was used to optimize the LS-SVM parameter values to overcome its sensitivity to changes in its parameter
values. These parameters are the regularization factor γ and Gaussian Kernel function σ; and ii) second stage:
classification. This stage consisting of two important stages: training stage then followed by testing stage.
The goal of this stage is classifying the Type-II-DM patients into one of two classes Healthy and DM.

Figure 1. A block diagram of the proposed algorithm COA-LS-SVM

The following is the pseudo code-4 that explain the proposed algorithm COA-LS-SVM in details:

The pseudo code-4. Proposed Algorithm COA-LS-SVM

Enter the data set of n data points, {xk , yk }N
k=1, where xi is the i
th input vector and y ∈ R is
i
th
the conforming i target with values {-1,+1}.
For each enter data point, randomly generate weights.
For each enter data point, randomly set the initial bias b and error e.
Calculate the optimal values of γ and σ using the pseudo code-2.
Calculate the optimal values of (w, b, e) for the objective function using (15) and (16).
Calculate the number of support vectors ∝ using (20).
Classify any new data point using (17) and Kernel function RBF using (22).
Repeat till stopping criteria is met, usually until the maximum number of iterations is
reached.

3.2. Data set

PIDD used in this research was collected from the machine learning database at UCI repository and
all the details about it are available in [46]. The data set consists of 768 cases whom were at least 21 years
old. Table 1 and Figure 2 are summarized the information and features about this data set.

Table 1. Information of data set

Data set No. of cases Feature Input Output classes
Pima 768 8 2
Healthy cases 𝐷𝑀 cases
500 268

A proposed model for diabetes mellitus classification … (Baydaa Sulaiman Bahnam)

1170  ISSN: 2252-8938

Figure 2. Features data set

4. EXPERIMENTAL RESULTS
The input of the COA is 768 cases of PIDD. In the search domain, these cases are randomly created
for 100 iterations. The output from the first stage COA of the proposed algorithm is the optimal values of
LS-SVM parameters which are γ=100 and σ=0.5. These optimal parameters used with the second stage of the
proposed algorithm LS-SVM classifier and RBF kernel function (22), in order to find the optimal hyperplane
that detaches the search area into two classes (Healthy, DM) by calculating the optimal values of (w, b, e) in
the objective function (15) and (16).
Accuracy metric was used to evaluate the performance of the proposed method [47]–[50]:
𝑇𝑃+𝑇𝑁
𝐴𝑐𝑐𝑢𝑟𝑎𝑐𝑦 = (23)
𝑇𝑃+𝑇𝑁+𝐹𝑃+𝐹𝑁

where: TP = True Positive denote the numbers of cases correctly diagnosed

TN = True Negative
FN = False Negative denote the numbers of cases incorrectly diagnosed
FP = False Positive
where the records with healthy label denotes positive cases while DM label denotes negative ones.
The proposed COA-LS-SVM algorithm is validated using the k-fold cross validation (K-Fold CV)
method for getting the best average accuracy value. K-Fold CV divides data into 𝐾 folds. At each iteration,
one-fold (𝐾) is used as test data set while training data set is resided folds (K1) in 𝐾 experiments [51], [52].
In this work, the value of 𝐾 = 10 folds, nine data sets for training and one for testing, then repeat this process
ten times until all data has been evaluated. Figure 3 illustrates the 10-Fold CV.
The testing average accuracy value of LS-SVM is 98.811% using the kernel function RBF for 10
times iteration. Table 2 shows the testing accuracy value for each 10-Fold CV. The performance of the
proposed model COA-LS-SVM has been compared with the models of other works using the PIDD database.
The main objective is to diagnose whether the patient is diabetic or not using this data. It is appropriate to
analyze and evaluate the result of the proposed model with other works since the past 10 years, using the
accuracy scale of classification. Table 3 shows the comparison and analysis of the proposed model with the
previous works selected based on the classification of accuracy; in addition to the number of cases used in
each study.

Int J Artif Intell, Vol. 11, No. 3, September 2022: 1164-1174

Int J Artif Intell ISSN: 2252-8938  1171

Figure 3. 10 fold cross validation

Table 2. Accuracy value for each 10-fold CV
Fold No. Accuracy value
Fold 1 95.953%
Fold 2 96.963%
Fold 3 98.837%
Fold 4 99.98%
Fold 5 98.981%
Fold 6 99.678%
Fold 7 98.9359%
Fold 8 99.99%
Fold 9 99.99%
Fold 10 98.81%
Average 98.811%

Figure 4 depictes the comparison of the proposed model with the previous approches that used
PIDD. This graph denotes that the this work has outdone previous approches. The highest average
classification accuracy using proposed model COA-LS-SVM is 98.811% which has outperformed the other
models.

Table 3. A comparative study of related research works for average classification accuracy of PIDD
Algorithm Accuracy No. of cases
PCA, Kmeans algorithm [53] 72% 768 cases
RB-bayes algorithm [15] 72.9% 768 cases
Cuckoo-fuzzy-KNN [22] 74.8% 768 cases
DTDN [16] 76% 768 cases
SVM [54] 78% 460 cases
LR, GB [11] 79% 768 cases
Naïve Bayes [55] 79.56% 768 cases
GB [13] 81.25% 768 cases
J48graft [17] 81.33% 768 cases
Multi-layer feed-forward neural network [23] 82% 768 cases
Hyper MLP [18] 82.4% 768 cases
GA-BPN [20] 84.713% 392 cases
SOM [14] 85% 768 cases
Neural network with genetic algorithm [56] 87.46% 768 cases
LDA-MWSVM [57] 89.74% 768 cases
AMMLP [19] 89.93% 768 cases
K-means and DT [58] 90.03% 768 cases
A modified mayfly-SVM [12] 94.5% 768 cases
Fuzzy, DT, ACO and ANN model [21] 95.852% 247 cases
The proposed algorithm 98.811% 768 cases

Figure 4. The classification accuracies of proposed model COA-LS-SVM and other models
A proposed model for diabetes mellitus classification … (Baydaa Sulaiman Bahnam)
1172  ISSN: 2252-8938

5. CONCLUSION
The diagnosis of Type-II-DM has a significant impact on raising health awareness in the
community. Therefore, the proposed models for diagnosing this disease can help the practitioners and the
patient to avoid its danger, reduce its complications and prevent it. To improve the diagnostic performance of
Type-II-DM disease more efficiently, an effective model based on COA-LS-SVM approach has been
proposed. The COA algorithm was used in the first stage to optimize the parameters of LS-SVM to overcome
its problem which is very sensitive when its parameter values are changed. Then LS-SVM classifier was
employed to classify Type-II-DM. Optimizing LS-SVM parameters using COA algorithm can ensure the
robustness and effectiveness of the proposed model by finding for optimal values instead of trial and error, as
well as making the classification more accurate and in less time. For verifying the efficiency of the proposed
model, experiments were performed on the PIDD dataset by detecting Type-II-DM and comparing the
accuracy of the model with the others models. The average accuracy of the proposed model was 98.811%
which significantly outperformed the others previous models implemented on PIDD. And as a work for the
near future, COA can be as an optimization technique and hybridized with other classification algorithms.
Also, other evaluation parameters can be applied as well as other kernel functions.

REFERENCES
[1] “IDF diabetes atlas,” International Diabetes Federation. 10th Ed. 2021. (Accessed: Sep. 2, 2021). [Online]. Available:
https://www.diabetesatlas.org/.
[2] “Diabetes,” World Health Organization. (Accessed Sep. 2, 2021). [Online]. Available: https://www.who.int/news-room/
fact-sheets/detail/diabetes.
[3] “2. Classification and diagnosis of diabetes: standards of medical care in diabetes - 2021,” Diabetes Care, vol. 44, pp. 15–33, Jan.
2021, doi: 10.2337/dc21-S002.
[4] O. AlShorman, M. S. Masadeh, and B. AlShorman, “Mobile health monitoring based studies for diabetes mellitus: a review,”
Bull. Electr. Eng. Informatics, vol. 10, no. 3, pp. 1405–1414, Jun. 2021, doi: 10.11591/eei.v10i3.3019.
[5] A. L., H. Singhal, I. Dwivedi, and P. Ghuli, “Diabetic retinopathy classification using deep convolutional neural network,”
Indones. J. Electr. Eng. Comput. Sci., vol. 24, no. 1, pp. 208–216, Oct. 2021, doi: 10.11591/ijeecs.v24.i1.pp208-216.
[6] R. Sofiana and S. Sutikno, “Optimization of backpropagation for early detection of diabetes mellitus,” Int. J. Electr. Comput.
Eng., vol. 8, no. 5, pp. 3232–3237, Oct. 2018, doi: 10.11591/ijece.v8i5.pp3232-3237.
[7] J. Singh, “Centers for disease control and prevention,” Indian Journal of Pharmacology, vol. 36, no. 4. Qeios, pp. 268–269, Feb.
2004, doi: 10.1097/jom.0000000000001045.
[8] M. M. Rosli, N. S. M. Yusop, and A. S. Fazuly, “Design of meal intake prediction for gestational diabetes mellitus using genetic
algorithm,” IAES Int. J. Artif. Intell., vol. 9, no. 4, pp. 591–599, 2020, doi: 10.11591/ijai. v9.i4.
[9] S. M. Rosli, M. M. Rosli, and R. Nordin, “A mapping study on blood glucose recommender system for patients with gestational
diabetes mellitus,” Bull. Electr. Eng. Informatics, vol. 8, no. 4, pp. 1489–1495, Dec. 2019, doi: 10.11591/eei.v8i4.1633.
[10] L. Syafaah, S. Basuki, F. Dwi Setiawan Sumadi, A. Faruq, and M. H. Purnomo, “Diabetes prediction based on discrete and
continuous mean amplitude of glycemic excursions using machine learning,” Bull. Electr. Eng. Informatics, vol. 9, no. 6,
pp. 2619–2629, Aug. 2020, doi: 10.11591/eei.v9i6.2387.
[11] R. Patil and S. Tamane, “A comparative analysis on the evaluation of classification algorithms in the prediction of diabetes,” Int.
J. Electr. Comput. Eng., vol. 8, no. 5, p. 3966, Oct. 2018, doi: 10.11591/ijece.v8i5.pp3966-3975.
[12] R. Patil, S. Tamane, S. A. Rawandale, and K. Patil, “A modified mayfly-SVM approach for early detection of type 2 diabetes
mellitus,” Int. J. Electr. Comput. Eng., vol. 12, no. 1, pp. 524–533, Feb. 2022, doi: 10.11591/ijece.v12i1.pp524-533.
[13] M. Panda, D. P. Mishra, S. M. Patro, and S. R. Salkuti, “Prediction of diabetes disease using machine learning algorithms,” IAES
Int. J. Artif. Intell., vol. 11, no. 1, pp. 284–290, Mar. 2022, doi: 10.11591/ijai.v11.i1.pp284-290.
[14] S. A. Diwan Alalwan, “Diabetic analytics: proposed conceptual data mining approaches in type 2 diabetes dataset,” Indones. J.
Electr. Eng. Comput. Sci., vol. 14, no. 1, p. 88, Apr. 2019, doi: 10.11591/ijeecs.v14.i1.pp88-95.
[15] R. Rajni and A. Amandeep, “RB-bayes algorithm for the prediction of diabetic in Pima Indian dataset,” Int. J. Electr. Comput.
Eng., vol. 9, no. 6, pp. 4866–4872, Dec. 2019, doi: 10.11591/ijece.v9i6.pp4866-4872.
[16] M. R. Bozkurt, N. Yurtay, Z. Yilmaz, and C. Sertkaya, “Comparison of different methods for determining diabetes,” Turkish J.
Electr. Eng. Comput. Sci., vol. 22, no. 4, pp. 1044–1055, 2014, doi: 10.3906/elk-1209-82.
[17] R. M. Rahman and F. Afroz, “Comparison of various classification techniques using different data mining tools for diabetes
diagnosis,” J. Softw. Eng. Appl., vol. 06, no. 03, pp. 85–97, 2013, doi: 10.4236/jsea.2013.63013.
[18] M. Khashei, S. Eftekhari, and J. Parvizian, “Diagnosing diabetes type II using a soft intelligent binary classification model,” Rev.
Bioinforma. Biometrics, vol. 1, no. 1, pp. 9–23, 2012.
[19] A. Marcano-Cedeño, J. Torres, and D. Andina, “A prediction model to diabetes using artificial metaplasticity,” in Lecture Notes
in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 6687,
no. 2, Springer Berlin Heidelberg, 2011, pp. 418–425, doi: 10.1007/978-3-642-21326-7_45.
[20] A. G. Karegowda, A. S. Manjunath, and M. A. Jayaram, “Application of genetic algorithm optimized neural network connection
weights for medical diagnosis of PIMA Indians diabetes,” Int. J. Soft Comput., vol. 2, no. 2, pp. 15–23, May 2011,
doi: 10.5121/ijsc.2011.2202.
[21] M. Fiuzy, A. Qarehkhani, J. Haddadnia, J. Vahidi, and H. Varharam, “Introduction of a method to diabetes diagnosis according to
optimum rules in fuzzy systems based on combination of data mining algorithm (DT), evolutionary algorithms (ACO) and
artificial neural networks (NN),” J. Math. Comput. Sci., vol. 06, no. 04, pp. 272–285, May 2013, doi: 10.22436/jmcs.06.04.03.
[22] R. Haritha, D. S. Babu, and P. Sammulal, “A hybrid approach for prediction of type-1 and type-2 diabetes using firefly and
cuckoo search algorithms,” Int. J. Appl. Eng. Res., vol. 13, no. 2, pp. 896–907, 2018.
[23] Y. Zhang, Z. Lin, Y. Kang, R. Ning, and Y. Meng, “A feed-forward neural network model for the accurate prediction of diabetes
mellitus,” Int. J. Sci. Technol. Res., vol. 7, no. 8, pp. 151–155, 2018.

Int J Artif Intell, Vol. 11, No. 3, September 2022: 1164-1174

Int J Artif Intell ISSN: 2252-8938  1173

[24] I. Obagbuwa, “Swarm intelligence algorithms and applications to real-world optimization problems: a survey,” Int. J. Simul. Syst.
Sci. Technol., May 2018, doi: 10.5013/ijssst.a.19.02.05.
[25] S. Selvaraj and E. Choi, “Survey of swarm intelligence algorithms,” in Proceedings of the 3rd International Conference on
Software Engineering and Information Management, Jan. 2020, pp. 69–73, doi: 10.1145/3378936.3378977.
[26] B. S. B. Alyas, “A particle swarm optimization algorithm based classifier for medical data set,” Int. J. Enhanc. Res. Sci. Technol.
Eng., vol. 4, no. 4, pp. 1–6, 2015.
[27] J. Pierezan and L. Dos Santos Coelho, “Coyote optimization algorithm: a new metaheuristic for global optimization problems,” in
2018 IEEE Congress on Evolutionary Computation (CEC), Jul. 2018, pp. 1–8, doi: 10.1109/CEC.2018.8477769.
[28] A. Fathy, M. Al-Dhaifallah, and H. Rezk, “Recent coyote algorithm-based energy management strategy for enhancing fuel
economy of hybrid FC/battery/SC system,” IEEE Access, vol. 7, pp. 179409–179419, 2019,
doi: 10.1109/ACCESS.2019.2959547.
[29] S. Mirjalili, S. M. Mirjalili, and A. Lewis, “Grey wolf optimizer,” Adv. Eng. Softw., vol. 69, pp. 46–61, Mar. 2014,
doi: 10.1016/j.advengsoft.2013.12.007.
[30] Z. Yuan, W. Wang, H. Wang, and A. Yildizbasi, “Developed coyote optimization algorithm and its application to optimal
parameters estimation of PEMFC model,” Energy Reports, vol. 6, pp. 1106–1117, Nov. 2020, doi: 10.1016/j.egyr.2020.04.032.
[31] T. T. Nguyen, T. T. Nguyen, N. A. Nguyen, and T. L. Duong, “A novel method based on coyote algorithm for simultaneous
network reconfiguration and distribution generation placement,” Ain Shams Eng. J., vol. 12, no. 1, pp. 665–676, Mar. 2021,
doi: 10.1016/j.asej.2020.06.005.
[32] G. W. Chang and N. Cong Chinh, “Coyote optimization algorithm-based approach for strategic planning of photovoltaic
distributed generation,” IEEE Access, vol. 8, pp. 36180–36190, 2020, doi: 10.1109/ACCESS.2020.2975107.
[33] E. M. Abdallah, M. I. El Sayed, M. M. Elgazzar, and A. A. Hassan, “Coyote multi-objective optimization algorithm for optimal
location and sizing of renewable distributed generators,” Int. J. Electr. Comput. Eng., vol. 11, no. 2, pp. 975–983, Apr. 2021,
doi: 10.11591/ijece.v11i2.pp975-983.
[34] D. Xiao, H. Li, and X. Sun, “Coal classification method based on improved local receptive field-based extreme learning machine
algorithm and visible–infrared spectroscopy,” ACS Omega, vol. 5, no. 40, pp. 25772–25783, Oct. 2020,
doi: 10.1021/acsomega.0c03069.
[35] J. Arfaoui, H. Rezk, M. Al-Dhaifallah, M. N. Ibrahim, and M. Abdelkader, “Simulation-based coyote optimization algorithm to
determine gains of PI controller for enhancing the performance of solar PV water-pumping system,” Energies, vol. 13, no. 17,
Aug. 2020, doi: 10.3390/en13174473.
[36] A. Abaza, R. A. El-Sehiemy, K. Mahmoud, M. Lehtonen, and M. M. F. Darwish, “Optimal estimation of proton exchange
membrane fuel cells parameter based on coyote optimization algorithm,” Appl. Sci., vol. 11, no. 5, p. 2052, Feb. 2021,
doi: 10.3390/app11052052.
[37] J. A. K. Suykens and J. Vandewalle, “Least squares support vector machine classifiers,” Neural Process. Lett., vol. 9, no. 3, pp.
293–300, 1999, doi: 10.1023/A:1018628609742.
[38] M. Farhadi and N. Mollayi, “Application of the least square support vector machine for point-to-point forecasting of the PV
power,” Int. J. Electr. Comput. Eng., vol. 9, no. 4, pp. 2205–2211, Aug. 2019, doi: 10.11591/ijece.v9i4.pp2205-2211.
[39] Z. M. Yasin, N. A. Salim, N. F. A. Aziz, Y. M. Ali, and H. Mohamad, “Long-term load forecasting using grey wolf optimizer -
least-squares support vector machine,” IAES Int. J. Artif. Intell., vol. 9, no. 3, pp. 417–423, Sep. 2020,
doi: 10.11591/ijai.v9.i3.pp417-423.
[40] L. Chen, L. Duan, Y. Shi, and C. Du, “PSO_LSSVM prediction model and its MATLAB implementation,” IOP Conf. Ser. Earth
Environ. Sci., vol. 428, no. 1, Jan. 2020, doi: 10.1088/1755-1315/428/1/012089.
[41] Z. Mat Yasin, N. Ashida Salim, N. F. Ab Aziz, H. Mohamad, and N. Ab Wahab, “Prediction of solar irradiance using grey wolf
optimizer-least-square support vector machine,” Indones. J. Electr. Eng. Comput. Sci., vol. 17, no. 1, pp. 10–17, Jan. 2020,
doi: 10.11591/ijeecs.v17.i1.pp10-17.
[42] A. Youssef, A. Youssef Ali Amer, N. Caballero, and J.-M. Aerts, “Towards online personalized-monitoring of human thermal
sensation using machine learning approach,” Appl. Sci., vol. 9, no. 16, p. 3303, Aug. 2019, doi: 10.3390/app9163303.
[43] W. Ma and H. Liu, “Classification method based on the deep structure and least squares support vector machine,” Electron. Lett.,
vol. 56, no. 11, pp. 538–541, May 2020, doi: 10.1049/el.2019.3776.
[44] Z. Tian, “Backtracking search optimization algorithm-based least square support vector machine and its applications,” Eng. Appl.
Artif. Intell., vol. 94, Sep. 2020, doi: 10.1016/j.engappai.2020.103801.
[45] M. Sivaram et al., “An optimal least square support vector machine based earnings prediction of blockchain financial products,”
IEEE Access, vol. 8, pp. 120321–120330, 2020, doi: 10.1109/ACCESS.2020.3005808.
[46] “UCI Repository,” kaggle.com. (Accessed: Sept. 2, 2018). [Online]. Available: https://www.kaggle.com/uciml/pima-indians-
diabetes-database.
[47] K. S. Y. Pande, D. G. H. Divayana, and G. Indrawan, “Comparative analysis of naïve bayes and KNN on prediction of forex price
movements for GBP/USD currency at time frame daily,” J. Phys. Conf. Ser., vol. 1810, no. 1, Mar. 2021, doi: 10.1088/1742-
6596/1810/1/012012.
[48] M. O. Adebiyi, M. O. Arowolo, and O. Olugbara, “A genetic algorithm for prediction of RNA-seq malaria vector gene expression
data classification using SVM kernels,” Bull. Electr. Eng. Informatics, vol. 10, no. 2, pp. 1071–1079, Apr. 2021,
doi: 10.11591/eei.v10i2.2769.
[49] C. Do, N. Q. Dam, and N. T. Lam, “Optimization of network traffic anomaly detection using machine learning,” Int. J. Electr.
Comput. Eng., vol. 11, no. 3, pp. 2360–2370, Jun. 2021, doi: 10.11591/ijece.v11i3.pp2360-2370.
[50] C.-Y. Lim, “Methods for evaluating the accuracy of diagnostic tests,” Cardiovasc. Prev. Pharmacother., vol. 3, no. 1, 2021,
doi: 10.36011/cpp.2021.3.e2.
[51] P. Ghosh, A. Karim, S. T. Atik, S. Afrin, and M. Saifuzzaman, “Expert cancer model using supervised algorithms with a LASSO
selection approach,” Int. J. Electr. Comput. Eng., vol. 11, no. 3, pp. 2631–2639, Jun. 2021, doi: 10.11591/ijece.v11i3.pp2631-
2639.
[52] F. R. Lumbanraja, E. Fitri, Ardiansyah, A. Junaidi, and R. Prabowo, “Abstract classification using support vector machine
algorithm (case study: abstract in a computer science journal),” J. Phys. Conf. Ser., vol. 1751, no. 1, p. 12042, Jan. 2021,
doi: 10.1088/1742-6596/1751/1/012042.
[53] R. N. Patil and S. Tamane, “A novel scheme for predicting type 2 diabetes in women: using kmeans with PCA as dimensionality
reduction,” Int. J. Comput. Eng. Appl., pp. 76–87, 2017.
[54] V. A. Kumari and R. Chitra, “Classification of diabetes disease using support vector machine,” Int. J. Eng. Res. Appl., vol. 3,
no. 2, pp. 1797–1801, 2013.

A proposed model for diabetes mellitus classification … (Baydaa Sulaiman Bahnam)

1174  ISSN: 2252-8938

[55] A. Iyer, J. S, and R. Sumbaly, “Diagnosis of diabetes using classification mining techniques,” Int. J. Data Min. Knowl. Manag.
Process, vol. 5, no. 1, pp. 1–14, Jan. 2015, doi: 10.5121/ijdkp.2015.5101.
[56] S. Mohammad, H. Dadgar, and M. Kaardaan, “A hybrid method of feature selection and neural network with genetic algorithm to
predict diabetes,” Int. J. Mechatronics, Electr. Comput. Technol., vol. 7, no. 24, pp. 3397–3404, 2017.
[57] D. Çalişir and E. Doğantekin, “An automatic diabetes diagnosis system based on LDA-wavelet support vector machine
classifier,” Expert Syst. Appl., vol. 38, no. 7, pp. 8311–8315, Jul. 2011, doi: 10.1016/j.eswa.2011.01.017.
[58] W. Chen, S. Chen, H. Zhang, and T. Wu, “A hybrid prediction model for type 2 diabetes using K-means and decision tree,” in
2017 8th IEEE International Conference on Software Engineering and Service Science (ICSESS), Nov. 2017, pp. 386–390,
doi: 10.1109/ICSESS.2017.8342938.

BIOGRAPHIES OF AUTHORS

Baydaa Sulaiman Bahnam is an Assistant Professor in Department of Software,

College of Computer Sciences and Mathematics, University of Mosul, Iraq since 2015. She
received her M.Sc. Degrees in Computer Sciences in the same College from University of
Mosul, Iraq, in 2006. Her research interests include the field of, artificial intelligence, swarm
intelligence, machine learning methods, classification, image processing, evolutionary
algorithms and data mining. She has a research gate account under the name Baydaa Sulaiman
Bahnam. She can be contacted at email: [email protected].

Suhair Abd Dawwod is currently lecturer in Department of Management

Information Systems College of Administration and Economics, University of Mosul, since
2001. I've got my B.Sc. and M.Sc. in Computer Science from the University of Mosul, Iraq in
1992 and 2001, respectively. My research interests include neural networks and artificial
intelligence, database, data mining, machine learning, computer networks and security. She
can be contacted at email: [email protected].

Int J Artif Intell, Vol. 11, No. 3, September 2022: 1164-1174