IAES International Journal of Artificial Intelligence (IJ-AI)

Vol. 14, No. 1, February 2025, pp. 213~221

ISSN: 2252-8938, DOI: 10.11591/ijai.v14.i1.pp213-221  213

Microarray gene expression classification: dwarf mongoose

optimization with deep learning

Shyamala Gowri Balaraman1, Anu H. Nair1, Sanal Kumar2

1
Department of Computer Science and Engineering, Faculty of Engineering and Technology, Annamalai University, Annamalai Nagar,
India
2
Department of Computer Science, Rajeswari Vedachalam Government Arts College, Chengalpattu, India

Article Info ABSTRACT

Article history: The deoxyribonucleic acid (DNA) microarray model holds significant
promise for revealing expression data from thousands of genes. It serves as a
Received Jan 29, 2024 valuable tool for investigating gene expressions in diverse biological research
Revised Aug 6, 2024 fields. This study explores advancements in gene selection for cancer
Accepted Aug 30, 2024 detection through artificial intelligence, with a focus on the challenge of
extracting pertinent information from vast databases. The application of deep
learning architecture in detecting chronic diseases and aiding medical
Keywords: decision-making has proven effective across various domains. Therefore, this
study designs an enhanced microarray gene expression classification by
Convolutional autoencoder utilizing a dwarf mongoose optimization with deep learning (MGEXC-
Deep learning DMODL) approach. The MGEXC-DMODL approach intends to classify the
Deoxyribonucleic acid microarray gene expression (MGE). For this, the MGEXC-DMODL
Dwarf mongoose optimization technique initially applies the wiener filtering (WF) technique to eradicate the
Microarray gene expression noise. In addition, the MGEXC-DMODL technique employs a deep residual
shrinkage network (DRSN) to learn feature vectors. Meanwhile, the
convolutional autoencoder (CAE) model was executed for identifying and
classifying the MGE data. Furthermore, the dwarf mongoose optimization
(DMO)-based hyperparameter tuning is performed to enhance the detection
outcomes of the CAE model. The investigational evaluation of the MGEXC-
DMODL model is validated using a benchmark database. The comprehensive
comparison outcome highlighted the betterment of the MGEXC-DMODL
model over recent approaches.
This is an open access article under the CC BY-SA license.

Corresponding Author:
Shyamala Gowri Balaraman
Research Scholar, Department of Computer Science and Engineering, Annamalai University
Annamalai Nagar, Chidambaram 608002, India
Email: [email protected]

1. INTRODUCTION
The microarray gene expression (MGE) data classification problem midpoints around the task of
precisely classifying biological samples dependent upon their profiles of gene expression [1]. The microarray
model permits scientists to evaluate many gene expression levels together, offering a wealth of data that can be
vital for recognizing diseases, finding biomarkers, and increasing targeted treatments [2]. However, this wealth
of information also delivers an important analytical and computational task. The problematic report contains
developing strong and effectual classification methods that can distinguish among samples like healthy and
unhealthy persons, based on their gene expression information [3]. In this situation, the main tasks contain
feature selection; dealing with high-dimensional data, and attaining great classification accuracy while securing

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

214  ISSN: 2252-8938

biological interpretability. Precise gene expression data classification is very essential for furthering our
consideration of difficult diseases, allowing initial diagnosis, and directing personalized treatment plans [4].
Microarray cancer data study is a vital study area across various fields like machine learning (ML),
pattern recognition, statistics, computational biology, and other associated areas. It plays a vital part in
recognition, analysis, and cancer treatment [5]. The studies are nowadays targeting the improvement of
existence rates in cancer patients by developing the process and knowledge of checking and treatment [6]. The
foremost trouble with microarray dataset identification arises from numerous issues like shortage of enough
samples, imbalanced class, noisy data, and high trouble of feature dimensionality that managed to be difficult
to diagnose and have outcomes of wrong classification. Several research works associated with dual-class data
classification of microarray cancer have been conducted [7]. Classifying multiclass data of microarray is still
an open research area due to an outcome of tasks in class imbalance. Classes with a tiny amount of models
have been generally ignored due to the bias of many methods near classes having more amount of elements
[8]. ML models are commonly used in resolving various difficult real issues and have been verified to be
effective in examining gene expression data. MGE data classification with deep learning (DL) is an innovative
technique that connects the power of neural networks to classify biological samples precisely based on their
MGE profiles [9]. In this procedure, high-dimensional gene expression data is changed into a plan that will be
appropriate for DL, and convolutional neural network (CNN) molecular basis of illnesses and allow
applications in precision medicine, biomarker identification, and drug discovery [10].
This study designs an enhanced microarray gene expression classification by utilizing a dwarf
mongoose optimization with deep learning (MGEXC-DMODL) approach. The MGEXC-DMODL approach
intends to classify the MGE. For this, the MGEXC-DMODL technique initially applies the Wiener filtering
(WF) technique to eradicate the noise. In addition, the MGEXC-DMODL technique employs a deep residual
shrinkage network (DRSN) to learn feature vectors. Meanwhile, the convolutional autoencoder (CAE) model
was executed for identifying and classifying the MGE data. Furthermore, the dwarf mongoose optimization
(DMO)-based hyperparameter tuning is performed to enhance the detection outcomes of the CAE model. The
investigational evaluation of the MGEXC-DMODL model is validated using a benchmark database.
The remaining sections of the article are arranged as: section 2 illustrates the related works. Section 3
portrays the proposed model. Then, section 4 elaborates on the experimental validation and section 5 completes
the work.

2. RELATED WORKS
Saheed [11] intended to develop an ML–based approach to categorize acute myeloid and acute
lymphoblastic leukemia dependent upon MGE profiles. The authors utilized linear discriminant analysis
(LDA), Ada boost, logistic regression (LR), k-neighbor method, extreme randomized trees algorithm, ridge
classifier, gradient boosting, and random forest (RF). The principle component analysis (PCA) was employed
for dimensionality reduction. The authors utilize 2 various cross-validation processes due to they make higher-
accurate ability evaluations than prior approaches. Vaiyapuri et al. [12] designed an innovative red fox
optimizer with a deep learning-based microarray gene expression classification (RFODL-MGEC) technique.
This model targets increasing classification effectiveness by choosing suitable features. The RFODL-MGEC
method employs an innovative request for offer (RFO)-based feature selection (FS) technique for determining
optimum feature subsets. Additionally, the RFODL-MGEC method includes a bi-directional cascaded deep
neural network (BCDNN) for classifying data. The constraints executed in the BCDNN method could be tuned
by employing the chaos game optimizer (CGO) technique.
Rostami et al. [13] introduced an innovative social network investigation-based gene selection
technique. The developed technique the relevance maximization and redundancy minimization (mRMR)
model. Here, at every round, a supreme community was preferred continually. Ke et al. [14] considered a
swarm-optimizer-assisted filter-wrapper gene selection comprising 2 stages: The primary stage will be the filter
step that chooses small top-n percentages of genes and attains decreased information; later, the secondary stage
examines for the optimum gene subsets depend upon a wrapper system in the residual genes by employing a
swarm optimization related technique. Research by Bacha et al. [15], an innovative decreased computer-aided
diagnosis (CAD) technique was applied with the MATLAB (version R2016a) platform for categorizing the
four cancer subcategories. The outcomes of the experiment have been performed with 4 groups of baseline
data under the appearance of cancerous genes.
Pandit et al. [16] projected an effective and hybrid DL method for classifying molecular cancer with
the help of expression data to resolve these borders. The input data was pre-processed employing a scalable
range adaptive bilateral filter (BF). Subsequently, clustering has been accomplished by employing an enriched
binomial clustering technique. Followed by the data must be removed through the multifractal Brownian
motion (MBM) technique. Later, the significant features should be chosen by utilizing an improved cuckoo

Int J Artif Intell, Vol. 14, No. 1, February 2025: 213-221

Int J Artif Intell ISSN: 2252-8938  215

search optimizer (ICSO) method. Lastly, the data classification was executed employing a wavelet-based deep
convolutional neural network (DCNN). Hilal et al. [17] presented new feature subset selection (FSS) with
optimum adaptive neuro fuzzy inference system (OANFIS) for classifying gene expression. The main goal is
to identify as well as categorize the gene expression information. To achieve this, the approach develops an
enhanced improved grey wolf optimizer-based feature selection (IGWO-FS) technique for achieving optimum
feature subsets. Further, the OANFIS technique was exploited in the classification of genes and the
hyperparameter tuning of the adaptive neuro-fuzzy inference system (ANFIS) system can be modified by
applying a coyote optimization algorithm (COA).

3. THE PROPOSED METHOD

In this study, an enhanced MGEXC-DMODL approach is designed. The MGEXC-DMODL approach
intends to classify the presence of the MGE classification. To accomplish this, the MGEXC-DMODL method
encompasses pre-processing, feature extractor, classification, and tuning processes. Figure 1 depicts the
structure of the MGEXC-DMODL method.

Figure 1. Workflow of MGEXC-DMODL technique

Microarray gene expression classification: dwarf mongoose optimization … (Shyamala Gowri Balaraman)
216  ISSN: 2252-8938

3.1. Preprocessing
Initially, the MGEXC-DMODL technique applies the WF technique to eradicate the noise that exists
in it. WF is an efficient way to enhance the accuracy and quality of microarray images [18]. The microarray
technique includes the simultaneous analysis of thousands of biological samples, generating large images with
imperfections and inherent noise. In this context, WF is used for enhancing the signal-to-noise ratio, enhancing
the clarity and reducing artifacts of gene expressions. The WF efficiently sharpens the image by statistically
modeling the characteristics of noise and desired signal (microarray spots representing gene expressions),
which facilitates quantification and more accurate detection of gene expressions. This technology achieved
remarkable success in genomics and bioinformatics research, assisting in the extraction of biological data from
microarray images and contributing to advancement in understanding complicated cellular processes.

3.2. Feature extraction

The MGEXC-DMODL technique employs DRSN to learn feature vectors. The concept behind the
integration of the RSN with the deep residual network has resulted in the formulation of the DRSN model [19].
DRSN is a complex cascaded deep neural network (DNN) that employs a soft threshold function and attention
mechanism to filter out noisy data. The DRSN uses self‐attention modules to systematically select valuable
feature data while removing noise and ineffectual features, hence boosting the DNN capacity to extract valuable
feature data from the noise.
The fundamental unit of DRSN is the residual shrinkage building unit (RSBU). The DRSN consists
of one identity mapping, two‐batch normalization (𝐵𝑁), one soft threshold learning subnetwork, 2 activation
functions (Mish), and 2 convolutional layers (𝐶𝑜𝑛𝑣). There is a subnetwork in each segment and its role is to
independently learn a group of thresholds. Therefore, the threshold is not too large and guaranteed to be
positive. The feature map learns various thresholds; hence the above subnetwork is utilized as an attention
module, and the soft threshold is used to convert the observed invalid feature into zero, and relevant features
are retained. The soft thresholding removes the feature closer to 0 and retains the negative and positive features.
𝑥 − 𝑡ℎ𝑟 𝑥 > 𝑡ℎ𝑟
𝑦 = {0 − 𝑡ℎ𝑟 ≤ 𝑥 ≤ 𝑡ℎ𝑟 (1)
𝑥 + 𝑡ℎ𝑟 𝑥 < −𝑡ℎ𝑟
In (1), 𝑥 and 𝑦 are the input and output features and 𝑡ℎ𝑟 shows the thresholding function.
1 𝑥 > 𝑡ℎ𝑟
𝜕𝑦
= {0 − 𝑡ℎ𝑟 ≤ 𝑥 ≤ 𝑡ℎ𝑟 (2)
𝜕𝑥
1 𝑥 < −𝑡ℎ𝑟
In (2) is derived from (1), and after derivation, the soft thresholding becomes 0 or 1.

𝑠𝑜𝑓𝑡 = (𝑥, 𝑎) = 𝑠𝑖𝑔𝑛(𝑥 ) ∗ max{|𝑥| − 𝑡ℎ𝑟, 0} (3)

The soft thresholding is converted into (3), where 𝑠𝑖𝑔𝑛(𝑥) denotes the symbolic function. The soft
threshold is to independently attain the threshold range. The DNN has better outcomes in self‐learning, hence
the incorporation of soft threshold and DNN can effectively differentiate features from the irrelevant features.
The principle of the attention module includes facilitating the NN model to learn input factors independently
and allocate weights to them. This enables to assignment of computational resources to obtain essential
features, which results in better performance. The attention module includes a key‐value mapping process via
query operation of matrix-vector. This involves deriving the corresponding weight values and calculating the
similarity between dimension vectors, which undergo normalization through the Softmax function.
Consequently, the weight values are multiplied with a matrix dimensional vector and their summation is
attained for formulating the last attention matrix. If 𝐾 = 𝑉 = 𝑄, then it is represented as a self‐attention module
and is given as (4).

𝑄𝐾𝑇
𝐴𝑡𝑡𝑒𝑛𝑡𝑖𝑜𝑛(𝑄, 𝐾, 𝑉) = 𝑠𝑜𝑓𝑡max { } (4)
√𝑑𝑘

Where 𝑄 = (𝑞1 , 𝑞2 , 𝑞𝑙 ) ∈ 𝑅𝑛𝑥𝑑 , 𝐾 = (𝑘1 , 𝑘2 , 𝑘𝑙 ) ∈ 𝑅𝑛𝑥𝑑 , 𝑉 = (𝑣1 , 𝑣2 , 𝑣𝑛 ) ∈ 𝑅𝑛𝑥𝑑 , 𝑑 refers to the dimensional
of single vector, 𝑛 indicates the amount of input vectors that are attained by linear conversion of input matrix
𝑋. 𝑑𝑘 indicates the matrix with dimension 𝑘 to adjust the inner products. 𝐾 𝑇 denotes the transposition of 𝐾,
and the formula of the input sequence linear mapping process is given as (5):

Int J Artif Intell, Vol. 14, No. 1, February 2025: 213-221

Int J Artif Intell ISSN: 2252-8938  217

𝑄 = 𝑊𝑞 𝑋
{𝐾 = 𝑊𝑘 𝑋 (5)
𝑉 = 𝑊𝑣 𝑋

where the linear mapping parameter matrices 𝑤𝑞 , 𝑤𝑘 , and 𝑤𝑣 are self-learned in the training model.

3.3. Classification using convolutional autoencoder model

At this phase, the CAE model can be executed for identifying and classifying the MGE data. CAE
combines the benefits of convolution filtering in CNN with unsupervised pretraining of autoencoders [20].
Rather than the fully connected (FC) layer, the encoder has a convolution layer and the decoder has a
deconvolution layer in contrast to the topology for autoencoders. The deconvolution filter is an inverse version
of the convolution filter. Additionally, the deconvolution layer should be followed by the unpooling layer. The
unpooling process can be done by keeping the location of maximum value during pooling, which preserves the
value of that location while unpooling and zeroing the rest.
Spatial locality can be retained by incorporating the convolution function at all the neurons. Thus, for
the input matrix 𝑃, the encoder computes.

𝑒𝑖 = 𝜎(𝑃 ∗ 𝐹 𝑛 + 𝑏) (6)

In (6), 𝜎 indicates the activation function, 𝑏 is encoder bias,∗ signifies 2𝐷 convolution, and 𝐹 𝑛 represents 𝑛𝑡ℎ
2D convolutional filter. Zero padding applies input matrix 𝑃 for retaining spatial resolution. Next, the
reconstruction is attained by (7).

𝑧𝑖 = 𝜎(𝑒𝑖 ∗ F̃ n + 𝑏̃) (7)

In (7), F̃ n shows 𝑛𝑡ℎ 2𝐷 convolution filters in the decoder, 𝑧𝑖 designates the reconstruction of 𝑖 𝑡ℎ input and 𝑏
indicates bias of the decoder. Unsupervised pre-training is used in the network that minimizes the (8).

𝐸 (𝜃) = ∑𝑚
𝑖=1( 𝑥𝑖 − 𝑧𝑖 )
2
(8)

The FC layer and softmax classifier are added and the decoder part is removed at the end of the network after
unsupervised pretraining of the unpooling and deconvolution layers.

3.4. Dwarf mongoose optimization-based hyperparameter tuning

Eventually, the DMO-based parameter tuning method is executed for enhancing the recognition
outputs of the CAE method. Chen et al. [21] developed a DMO algorithm which is a population-based
metaheuristic model. This method splits the mongoose populace into 3 dissimilar groups such as babysitter,
scout, and alpha. Below the control of a female leader, the whole populace jointly feeds as an adhesive unit. If
the group of alpha flops to find food, an interchange happens among followers of the babysitter and alpha
groups. So, associates of the alpha group at the same time are involved in hunting actions while penetrating for
a sleeping mound. DMO needs only physically organized limits to decrease the difficulty of the system use.
When the associates of the alpha group have inadequate aptitudes, they will interchange followers of
babysitters, and alpha groups offer DMO the capability to uphold populace variety. The sleep mound device
can stop the algorithm from arriving at local goals.
− Initialize
Set the DMO’s mathematical method, as presented in (9).

𝑋1,1 𝑋1,2 … 𝑋1,𝑑−1 𝑋1,𝑑

𝑋2,1 𝑋2,1 … 𝑋2,𝑑−1 𝑋2,𝑑
𝑋= (9)
⋮ ⋮ 𝑋𝑖,𝑗 ⋮ ⋮
[𝑋𝑁,1 𝑋𝑁,2 … 𝑋𝑁,𝑑−1 𝑋𝑁,𝑑 ]

Whereas 𝑋𝑖,𝑗 denotes the location of the 𝑖𝑡ℎ mongoose in the 𝑗𝑡ℎ dimension; 𝑁 signifies the populace number;
𝑋 signifies the solution of the candidate and 𝑑 is the size of the problem. The mathematical method is displayed
in (10).

𝑋𝑖,𝑗 = 𝑢𝑛𝑖𝑓𝑟𝑛𝑑(𝑙𝑏, 𝑢𝑏, 𝑑 ) (10)

Microarray gene expression classification: dwarf mongoose optimization … (Shyamala Gowri Balaraman)
218  ISSN: 2252-8938

Here 𝑢𝑛𝑖𝑓𝑟𝑛𝑑 is employed to make evenly spread random numbers; 𝑙𝑏 and 𝑢𝑏 denote the upper and lower
limits, correspondingly; and 𝑑 signifies the dimension.
− Alpha group
The foraging direction of the dwarf mongoose is defined by the female leader, who is formed in the
group of alpha. The possibility of every female individual in the alpha group fetching a leader has been defined
by (11).

𝑓𝑖𝑡(𝑖)
𝛼 = ∑𝑛 (11)
𝑖−1 𝑓 𝑖𝑡(𝑖)

where 𝑓𝑖𝑡(𝑖) denotes the fitness output of the 𝑖𝑡ℎ individual; 𝑛 = 𝑁 − 𝑏𝑠; 𝑛 signifies the number of individuals
in the group of alpha; and 𝑏𝑠 represents the individuals count in the babysitter group.
The alpha females have preferred foraging ways, and their formulation is as (12):

𝑋𝑖+1 = 𝑋𝑖 + 𝑝 × 𝑝𝑒𝑒𝑝 × (𝑋𝑖 − 𝑋𝑘 ) (12)

Whereas 𝑋𝑖 signifies the position of the 𝑖𝑡ℎ individual; 𝑋𝑖+1 denotes the novel food source place; 𝑝 signifies
the random amount among [−1,1]; 𝑋𝑘 is an arbitrary individual in the alpha group and 𝑝𝑒𝑒𝑝 is set to 2. The
sleeping mound (SM) is the relaxing location of dwarf mongooses. Its expression is as (13):

𝑓𝑖𝑡(𝑖+1)−𝑓𝑖𝑡(𝑖)
𝑠𝑚𝑖 = max{|𝑓𝑖𝑡(𝑖+1),𝑓𝑖𝑡(𝑖)|} (13)

The mathematical method of the mean SM is as (14):

∑𝑛
𝑖=1 𝑠𝑚𝑖
𝜑= 𝑛
(14)

− Scout group
The separate followers of the group of scouts will not arrive at their preceding SM. This promises the
algorithm’s exploration capability. The SM mathematical formula is as (15):

⃗⃗ | 𝑖𝑓 𝜑𝑖 + 1 > 𝜑𝑖
𝑋𝑖 − 𝐶 × 𝑝 × 𝑟 × |𝑋𝑖 − 𝑀
𝑋𝑖+1 = { (15)
⃗⃗ |
𝑋𝑖 + 𝐶 × 𝑝 × 𝑟 × |𝑋𝑖 − 𝑀 𝑒𝑙𝑠𝑒

Here 𝑋𝑖+1 denotes the location of the subsequent SM; 𝐶 signifies the parameter that monitors the flexibility of
the mongoose populace.
− Babysitter group
The babysitter group dimension naturally consists of sub-ordinate individuals concerned for their
offspring where it is defined as dependent upon the dimension of population. This affects the system by
consistently declining the alpha group foraging possible over the period. Parameter 𝐿 changes the data
regarding foraging places for other followers. The babysitter fitness weight is fixed to 0, which safeguards the
average weight of the alpha group in the following iteration has been decreased therefore means the group
effort is delayed.
The DMO model grows a fitness function (FF) for attaining superior classifier results. It expresses a
positive numeral to imply the best output for the candidate's efficiency. Here, the lessening of the classifier
errors is measured as FF, as denoted in (18).

𝑓𝑖𝑡𝑛𝑒𝑠𝑠(𝑥𝑖 ) = 𝐶𝑙𝑎𝑠𝑠𝑖𝑓𝑖𝑒𝑟𝐸𝑟𝑟𝑜𝑟𝑅𝑎𝑡𝑒(𝑥𝑖 )
𝑁𝑜.𝑜𝑓 𝑚𝑖𝑠𝑐𝑙𝑎𝑠𝑠𝑖𝑓𝑖𝑒𝑑 𝑠𝑎𝑚𝑝𝑙𝑒𝑠
= × 100 (16)
𝑂𝑣𝑒𝑟𝑎𝑙𝑙 𝑠𝑎𝑚𝑝𝑙𝑒𝑠

4. EXPERIMENTAL VALIDATION
The outcome evaluation of the MGEXC-DMODL method can be examined utilizing three benchmark
datasets [22], such as breast, colon, and ovarian cancer. A comprehensive comparison result of the MGEXC-
DMODL method on the breast cancer dataset can be highlighted in Figure 2 [23]–[25]. These outcomes pointed
out that the CGRMD-MR-ANFIS, grid-based, Fuzzy c means, and CNN model has shown the least
performance. Meanwhile, the RF model gains slightly boosted outcomes. However, the MGEXC-DMODL

Int J Artif Intell, Vol. 14, No. 1, February 2025: 213-221

Int J Artif Intell ISSN: 2252-8938  219

technique demonstrates maximum performance with 𝑎𝑐𝑐𝑢𝑦 of 94.59%, 𝑝𝑟𝑒𝑐𝑛 of 94.12%, 𝑟𝑒𝑐𝑎𝑙 of 94.59%,
and 𝐹𝑠𝑐𝑜𝑟𝑒 of 94.02%.

Figure 2. Comparative result of the MGEXC-DMODL system under breast cancer dataset

A wide comparative analysis of the MGEXC-DMODL method with the colon cancer dataset can be
emphasized in Figure 3. These obtained findings indicate that the genetic algorithm (GA)-support vector
machine (SVM), GA- K-nearest neighbors (KNN), random+SVM, PCA-voting, logistic bootstrap (LogitBoot),
and RF methods have shown poorer performance. Meanwhile, the two-way clustering technique achieves
moderated increased outcomes. Nevertheless, the MGEXC-DMODL model reveals supreme performance with
an 𝑎𝑐𝑐𝑢𝑦 of 96.15%, 𝑝𝑟𝑒𝑐𝑛 of 92.86%, 𝑟𝑒𝑐𝑎𝑙 of 96.15%, and 𝐹𝑠𝑐𝑜𝑟𝑒 of 94.15%.

Figure 3. Comparative outcome of the MGEXC-DMODL method under colon cancer dataset

An extensive comparative result of the MGEXC-DMODL method at the ovarian cancer dataset can
be underscored in Figure 4. These accomplished findings denote that the linear SVM, RF, ensemble SVM,
common feature optimization (CFO)-LDA, laplace approximation (LAPO)-KNN, and gradient boosted
classifier (GBCO)-LR techniques get poorer performance. Similarly, the adaptive ant colony optimization
(AAO)-multi-layer perceptron (MLP) technique obtains moderated boosted outcomes. However, the MGEXC-
DMODL technique shows excellent performance with an 𝑎𝑐𝑐𝑢𝑦 of 95.31%, 𝑝𝑟𝑒𝑐𝑛 of 96.81%, 𝑟𝑒𝑐𝑎𝑙 of
95.31%, and 𝐹𝑠𝑐𝑜𝑟𝑒 of 95.89%. These outcomes confirmed the boosted performance of the MGEXC-DMODL
method under gene expression classification.

Microarray gene expression classification: dwarf mongoose optimization … (Shyamala Gowri Balaraman)
220  ISSN: 2252-8938

Figure 4. Comparative outcome of the MGEXC-DMODL model with ovarian cancer dataset

5. CONCLUSION
In this study, enhanced MGEXC-DMODL approach is designed. The MGEXC-DMODL technique
intends to classify the presence of the MGE classification. To accomplish this, the MGEXC-DMODL
technique initially applies the WF technique to eradicate the noise that exists in it. In addition, the MGEXC-
DMODL technique employs DRSN to learn feature vectors. Meanwhile, the CAE technique can be executed
for the identification and classification of MGE data. Furthermore, the DMO-based hyperparameter tuning is
performed to improve the recognition outcomes of the CAE algorithm. The comparative analysis of the
MGEXC-DMODL methodology highlighted the superior outcome of 94.59%, 96.15%, and 95.31% over recent
state of art approaches under benchmark datasets. The MGEXC-DMODL methodology encounters restrictions
in handling large datasets and warrants exploration in real-world clinical settings, prompting future
enhancement in disease detection via MGE classification.

REFERENCES
[1] S. Osama, H. Shaban, and A. A. Ali, “Gene reduction and machine learning algorithms for cancer classification based on microarray
gene expression data: A comprehensive review,” Expert Systems with Applications, vol. 213, 2023, doi:
10.1016/j.eswa.2022.118946.
[2] M. Abd-Elnaby, M. Alfonse, and M. Roushdy, “Classification of breast cancer using microarray gene expression data: A survey,”
Journal of Biomedical Informatics, vol. 117, 2021, doi: 10.1016/j.jbi.2021.103764.
[3] R. Tabares-Soto, S. Orozco-Arias, V. Romero-Cano, V. S. Bucheli, J. L. Rodríguez-Sotelo, and C. F. Jiménez-Varón, “A
comparative study of machine learning and deep learning algorithms to classify cancer types based on microarray gene expression
data,” PeerJ Computer Science, vol. 2020, no. 4, 2020, doi: 10.7717/peerj-cs.270.
[4] E. Alhenawi, R. Al-Sayyed, A. Hudaib, and S. Mirjalili, “Feature selection methods on gene expression microarray data for cancer
classification: A systematic review,” Computers in Biology and Medicine, vol. 140, 2022, doi: 10.1016/j.compbiomed.2021.105051.
[5] J. O. Agushaka, A. E. Ezugwu, O. N. Olaide, O. Akinola, R. A. Zitar, and L. Abualigah, “Improved dwarf mongoose optimization
for constrained engineering design problems,” Journal of Bionic Engineering, vol. 20, no. 3, pp. 1263–1295, 2023, doi:
10.1007/s42235-022-00316-8.
[6] S. Aruna and L. V. Nandakishore, “Empirical analysis of the effect of resampling on supervised learning algorithms in predicting
the types of lung cancer on multiclass imbalanced microarray gene expression data,” EAI/Springer Innovations in Communication
and Computing, pp. 15–27, 2022, doi: 10.1007/978-3-030-86165-0_2.
[7] N. Fernandez-Pozo et al., “PEATmoss (Physcomitrella Expression Atlas Tool): a unified gene expression atlas for the model plant
Physcomitrella patens,” Plant Journal, vol. 102, no. 1, pp. 165–177, 2020, doi: 10.1111/tpj.14607.
[8] L. Chen, L. Klebanov, A. Almudevar, C. Proschel, and A. Yakovlev, “A study of the correlation structure of microarray gene
expression data based on mechanistic modelling of cell population kinetics,” Statistical Modeling for Biological Systems: In Memory
of Andrei Yakovlev, pp. 47–61, 2020, doi: 10.1007/978-3-030-34675-1_3.
[9] M. Loey, M. W. Jasim, H. M. EL-Bakry, M. H. N. Taha, and N. E. M. Khalifa, “Breast and colon cancer classification from gene
expression profiles using data mining techniques,” Symmetry, vol. 12, no. 3, 2020, doi: 10.3390/sym12030408.
[10] Y. Wang, H. Wei, L. Song, L. Xu, J. Bao, and J. Liu, “Gene expression microarray data meta-analysis identifies candidate genes
and molecular mechanism associated with clear cell renal cell carcinoma,” Cell Journal, vol. 22, no. 3, pp. 386–393, 2020, doi:
10.22074/cellj.2020.6561.
[11] Y. K. Saheed, “Effective dimensionality reduction model with machine learning classification for microarray gene expression data,”
Data Science for Genomics, pp. 153–164, 2022, doi: 10.1016/B978-0-323-98352-5.00006-9.
[12] T. Vaiyapuri, Liyakathunisa, H. Alaskar, E. Aljohani, S. Shridevi, and A. Hussain, “Red fox optimizer with data-science-enabled
microarray gene expression classification model,” Applied Sciences, vol. 12, no. 9, 2022, doi: 10.3390/app12094172.
[13] M. Rostami, S. Forouzandeh, K. Berahmand, M. Soltani, M. Shahsavari, and M. Oussalah, “Gene selection for microarray data

Int J Artif Intell, Vol. 14, No. 1, February 2025: 213-221

Int J Artif Intell ISSN: 2252-8938  221

classification via multi-objective graph theoretic-based method,” Artificial Intelligence in Medicine, vol. 123, 2022, doi:
10.1016/j.artmed.2021.102228.
[14] L. Ke, M. Li, L. Wang, S. Deng, J. Ye, and X. Yu, “Improved swarm-optimization-based filter-wrapper gene selection from
microarray data for gene expression tumor classification,” Pattern Analysis and Applications, vol. 26, no. 2, pp. 455–472, 2023,
doi: 10.1007/s10044-022-01117-9.
[15] S. Bacha, O. Taouali, and N. Liouane, “Reduced CAD system for classifications of cancer types based on microarray gene
expression data,” 2022 IEEE 9th International Conference on Sciences of Electronics, Technologies of Information and
Telecommunications, SETIT 2022, pp. 133–137, 2022, doi: 10.1109/SETIT54465.2022.9875863.
[16] D. Pandit, J. Dhodiya, and Y. Patel, “Molecular cancer classification on microarrays gene expression data using wavelet-based deep
convolutional neural network,” International Journal of Imaging Systems and Technology, vol. 32, no. 6, pp. 2262–2280, 2022, doi:
10.1002/ima.22780.
[17] A. M. Hilal et al., “Feature subset selection with optimal adaptive neuro-fuzzy systems for bioinformatics gene expression
classification,” Computational Intelligence and Neuroscience, vol. 2022, 2022, doi: 10.1155/2022/1698137.
[18] R. Liu, Y. Li, H. Wang, and J. Liu, “A noisy multi-objective optimization algorithm based on mean and Wiener filters,” Knowledge-
Based Systems, vol. 228, 2021, doi: 10.1016/j.knosys.2021.107215.
[19] T. Han, Z. Zhang, M. Ren, C. Dong, X. Jiang, and Q. Zhuang, “Speech emotion recognition based on deep residual shrinkage
network,” Electronics, vol. 12, no. 11, 2023, doi: 10.3390/electronics12112512.
[20] C. Campbell and F. Ahmad, “Semi-supervised attention-augmented convolutional autoencoder for radar-based human activity
recognition,” SPIE Defense + Commercial Sensing, 2022, doi: 10.1117/12.2622366.
[21] S. Chen, Y. Zhou, and Q. Luo, “Hybrid adaptive dwarf mongoose optimization with whale optimization algorithm for extracting
photovoltaic parameters,” AIMS Energy, vol. 12, no. 1, pp. 84–118, 2024, doi: 10.3934/energy.2024005.
[22] Zhu et al. “Microarray datasets,” Weka ARFF format. [Online]. Available: http://csse.szu.edu.cn/staff/zhuzx/Datasets.html
[23] P. Mishra and N. Bhoi, “Cancer gene recognition from microarray data with manta ray based enhanced ANFIS technique,”
Biocybernetics and Biomedical Engineering, vol. 41, no. 3, pp. 916–932, 2021, doi: 10.1016/j.bbe.2021.06.004.
[24] A. El-Nabawy, N. El-Bendary, and N. A. Belal, “Epithelial ovarian cancer stage subtype classification using clinical and gene
expression integrative approach,” Procedia Computer Science, vol. 131, pp. 23–30, 2018, doi: 10.1016/j.procs.2018.04.181.
[25] S. K. Prabhakar and S. W. Lee, “An integrated approach for ovarian cancer classification with the application of stochastic
optimization,” IEEE Access, vol. 8, pp. 127866–127882, 2020, doi: 10.1109/ACCESS.2020.3006154.

BIOGAPHIES OF AUTHOR

Shyamala Gowri Balaraman is a Research Scholar in the Department of Computer

Science and Engineering, Annamalai University from 2021 January. She has published 2
research papers in international journals and conferences. Currently, her research projects
include image processing, machine learning, deep learning, and data science. She can be
contacted at email: [email protected].

Anu H. Nair is an Assistant Professor in the Department of Computer Science and

Engineering, Annamalai University from 2005. Her main research areas include image
processing in machine learning techniques. She has authored 70 research papers in international
journals and conferences, and two book chapters. Her current research projects include big data
analytics, biometric person identification, and medical image processing. She can be contacted
at email: [email protected].

Sanal Kumar is currently working as Assistant Professor in the P.G. Department

of Computer Science at R.V. Government Arts College, Chengalpattu, Tamilnadu, India (on
Deputation). He has authored 65 papers in international journals and conferences and holds life
membership in CSI, ISTE, and IAENG. His research interests span image processing, pattern
recognition, and wearable computing. He can be contacted at email: [email protected].

Microarray gene expression classification: dwarf mongoose optimization … (Shyamala Gowri Balaraman)