A Comparative Study of Enhanced Machine Learning

Algorithms for Brain Tumor Detection and Classification

This paper was downloaded from TechRxiv (https://www.techrxiv.org).

LICENSE

CC BY 4.0

SUBMISSION DATE / POSTED DATE

23-10-2021 / 26-10-2021

CITATION

GHOSH, ANKIT; KOLE, ALOK (2021): A Comparative Study of Enhanced Machine Learning Algorithms for
Brain Tumor Detection and Classification. TechRxiv. Preprint. https://doi.org/10.36227/techrxiv.16863136.v1

DOI

10.36227/techrxiv.16863136.v1
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 1

A Comparative Study of Enhanced Machine

Learning Algorithms for Brain Tumor Detection
and Classification
Ankit Ghosh, M.E. Student, School of Nuclear Studies and Application, Jadavpur University, Kolkata,
India, Alok Kole, Professor, Department of Electrical Engineering, RCC Institute of Information
Technology, Kolkata, India

 Index Terms—artificial intelligence, binary classification,

Abstract—The improvement of Artificial Intelligence (AI) and brain tumor classification, comparative analysis, machine
Machine Learning (ML) can help radiologists in tumor learning, MRI image, multi-class classification
diagnostics without invasive measures. Magnetic resonance
imaging (MRI) is a very useful method for diagnosis of tumors in I. INTRODUCTION
human brain. In this paper, brain MRI images have been
analyzed to detect the regions containing tumors and classify
these regions into three different tumor categories: meningioma, T HE applications of Artificial Intelligence (AI) and
Machine Learning (ML) have been expanding at an
exponential rate [1], [2], [3], [4], [5], [6], [7], [8], [9], [10].
glioma, and pituitary. This paper presents the implementation
and comparison of various enhanced ML algorithms for the There has been a substantial growth in the applications of AI
detection and classification of brain tumors. A brain tumor is the and ML techniques in various radiological imaging tasks [11],
growth of abnormal cells in the human brain. Brain tumors can [12]. AI and ML have been playing a major role in uncovering
be cancerous or non-cancerous. Cancerous or malignant brain
hidden insights into clinical decision making [13], [14]. ML is
tumors can be life threatening. Hence, detection and classification
of brain tumors at an early stage is extremely important. In this used in clinical practice including digital pathology and chest,
paper, enhanced ML algorithms have been implemented to brain, cardiovascular and abdominal imaging. Merging AI and
predict the presence or the absence of brain tumors using binary ML with the competency of medical practitioners, it is feasible
classification and to predict whether a patient has brain tumor or to enhance the productivity and potentially improve the
not and if he does, detect the type of brain tumor using multi- accuracy [15], [16], [17].
class classification. The dataset that has been used to perform the
It is essential to promote early diagnosis of brain tumors
binary classification task comprises of two types of brain MRI
images with tumor and without tumor. Here nine ML algorithms because they are the most common cause of cancer-related
namely, Support Vector Machine (SVM), Logistic Regression, K- deaths in children and people up to 40 years of age. Therefore,
Nearest Neighbor (KNN), Naïve Bayes (NB), Decision Tree (DT) it is necessary to devise strategies to accelerate early diagnosis
classifier, Random Forest classifier, XGBoost classifier, of brain tumors. An early diagnosis of brain tumor implies
Stochastic Gradient Descent (SGD) classifier and Gradient faster response in treatment, thereby increasing the surviving
Boosting classifier have been used to classify the MRI images. A
rates of patients. A process designed to automatically detect,
comparative analysis of the ML algorithms has been performed
based on a few performance metrics such as accuracy, recall, and locate and classify brain tumors is desirable. AI and ML have
precision, F1-score, AUC-ROC curve and AUC-PR curve. gained prominence in almost every field of decision-making
Gradient Boosting classifier has outperformed all the other and can be successfully implemented for the detection and
algorithms with an accuracy of 92.4%, recall of 94.4%, precision classification of brain tumors.
of 85%, F1-score of 89.5%, AUC-ROC of 97.2% and an AUC-PR The objective of this paper is to investigate the use of ML
of 91.4%. To address the multi-class classification problem, four
classification algorithms to detect the presence of brain tumors
ML algorithms namely, SVM, KNN, Random Forest classifier
and XGBoost classifier have been employed. In this case, the and also distinguish between different types of brain tumors
dataset that has been used consists of four types of brain MRI such as glioma, meningioma and pituitary tumors from brain
images with glioma tumor, meningioma tumor, and pituitary MRI images. A computer-aided classification method is more
tumor and with no tumor. The performances of the ML reliable for brain tumor diagnostics. The proposed scheme
algorithms have been compared based on accuracy, recall, involves a few steps including data collection, data pre-
precision and the F1-score. XGBoost classifier has surpassed all
processing (data labeling and image pre-processing),
the other algorithms in terms of accuracy, precision, recall and
F1-score. XGBoost has produced an accuracy of 90%, precision classification based on enhanced ML techniques and finally a
of 90%, and recall of 90% and F1-score of 90%. comparative analysis of the implemented models.

II. RELATED WORK

A lot of research work has been done in the field of
Artificial Intelligence (AI) and Machine Learning (ML)
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 2

application in the field of medical imaging. Noreen et al. [18] the four classes (normal, glioblastoma, sarcoma and metastatic
have proposed the use of two pre-trained deep learning models bronchogenic carcinoma tumors).
i.e. Inception-v3 and DensNet201 for developing a multi-level In [23] Rehman et al. have conducted three studies using
feature extraction and concatenation method for the early three architectures of convolutional neural networks (AlexNet,
detection of brain tumors and their classification. At first, they GoogLeNet, and VGGNet) to perform the classification of
have extracted the features from different Inception modules brain tumors such as meningioma, glioma, and pituitary. Then
from the pre-trained Inception-v3 model. Then they have they have explored the transfer learning techniques, i.e., fine-
passed those features to the softmax classifier to perform the tune and freeze using MRI slices of brain tumor dataset—
classification of the brain tumors. Secondly, they have used a Figshare. They have applied data augmentation techniques to
pre-trained DensNet201 to extract features from various the MRI images to generalize the results, increase the dataset
DensNet blocks. Then they have concatenated those features samples and reduce the chance of over-fitting. The proposed
and passed them to the softmax classifier to classify the brain fine-tune VGG16 architecture has attained the highest
tumors. The dataset that they have used comprised of three accuracy up to 98.69% in terms of classification and detection.
classes of brain tumors and it is available publicly. Their
proposed methodology has produced exceptional results and III. DATASETS
has outperformed all the existing state-of-the-art ML and Deep Two different datasets have been used:
Learning (DL) models for brain tumor detection and 1. Dataset-A (binary classification)
classification. 2. Dataset-B (multi-class classification)
In [19] Naik and Patel have used the decision tree
classification algorithm for the detection and classification of A. Dataset-A (binary classification)
brain tumor from MRI images. In the pre-processing step they Dataset-A has been used for binary classification. It
have used the median filtering process and texture feature comprises of 982 brain MRI images of patients with tumor
extraction technique has been used to extract the features. and 493 images with no tumor. Thus, a total of 1475 images
Their proposed model has exhibited improved efficiency in are present. A collection of 18 brain tumor images from
comparison to the traditional image mining methods. The Dataset-A are shown in Fig.1.
results that they have obtained have been compared with the
Naïve Bayesian classification algorithm. The decision tree
classification algorithm has achieved a precision of 100%,
Sensitivity of 93%, Specificity of 100% and Accuracy of 96%.
Tandel et al. [20] have proposed a transfer-learning-based
AI paradigm using a Convolutional Neural Network
(CNN) for brain tumor classification using MRI data. The
transfer-learning-based CNN model has been benchmarked
against six different ML classification algorithms, namely
Decision Tree, Linear Discrimination, Naive Bayes, Support
Vector Machine, K-nearest neighbour and Ensemble. Their
proposed model has proven to be very useful in multiclass
brain tumour grading and has yielded better results in Fig. 1. Collection of 18 brain tumor MRI images from Dataset-A
comparison to the other ML models.
Sarhan [21] has presented a computer-aided detection 15 images with no brain tumor from Dataset-A are shown in
(CAD) technique for the classification of brain tumors in MRI Fig.2.
images. The features from the brain MRI images have been
extracted by utilizing the Discrete Wavelet Transform (DWT).
The extracted features have then been applied to a CNN to
classify the input MRI image. His proposed approach has
produced an overall accuracy of 98.5%.
Mohsen et al. [22] in their research work have proposed the
development of a Deep Neural Network (DNN) classifier for
the classification of brain tumors on a dataset comprising of
66 brain MRI images of 4 types of brain tumors, namely,
normal, glioblastoma, sarcoma and metastatic bronchogenic
carcinoma tumors. They have combined the classifier with
DWT for feature extraction and principal components analysis
(PCA). The DNN classifier yielded extremely good results
with an average classification rate of 96.97%, average recall of
0.97, average precision of 0.97, average F-Measure of 0.97
Fig. 2. Collection of 15 brain MRI images with no tumor from Dataset-A
and average area under the ROC curve (AUC) of 0.984 of all
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 3

B. Dataset-B (multi-class classification)

Dataset-B has been used for multi-class classification. It
consists of 826 brain MRI images of glioma tumor, 822
images of meningioma tumor, 827 images of pituitary tumor
and 395 images with no tumor. Thus, a total of 2870 images
are present. A collection of 8 glioma tumor images from
Dataset-B are shown in Fig.3.

Fig. 6. Collection of 8 brain MRI images with no tumor from Dataset-B

The presence of a white spot as marked by the red circles in

Fig. 1, 3, 4, and 5 is an indication of the abnormal growth of
tissues in the human brain. There is an aggregation of
abnormal cells in some tissues of the brain in the above
Fig. 3. Collection of 8 glioma brain tumor MRI images from Dataset-B mentioned section. The tumors in these images are quite
critical. Manual classification of these images is rather
Fig.4 shows a collection of eight meningioma brain tumor difficult. So, ML algorithms have been employed for the
images from Dataset-B. efficient detection and classification of brain tumors from
these MRI images.

IV. STATE-OF-THE-ART MACHINE LEARNING ALGORITHMS

In this section some of the state-of-the-art ML algorithms
that can be used for the detection and classification of brain
tumors have been discussed.
Support Vector Machine (SVM) is a supervised ML
classification algorithm designed by a separative hyper-plane.
The main purpose of the SVM is to segregate the data in the
best possible way [24]. Therefore, SVM is a frontier which
best segregates the two classes. Logistic Regression is a
Fig. 4. Collection of 8 meningioma brain tumor MRI images from Dataset-B supervised ML algorithm which is used to predict a binary
outcome based on a set of independent variables. The main
A collection of eight pituitary brain tumor images from aim of logistic regression is to find the best fitting model to
Dataset-B are shown in Fig.5. describe the relationship between the outcome and a set of
predictor variables [25]. K-Nearest Neighbor (KNN) is a
supervised ML algorithm. It is used to solve binary
classification problems. KNN predicts whether a given data
point belongs to a particular class or the other by calculating
the distance between the given data point and the other points
[26]. The given data point belongs to that class whose data
points are nearest to it. K in KNN refers to the number of
points to be selected in the vicinity of the given data point.
Naïve Bayes (NB) is a supervised ML algorithm used mostly
for binary classification. It is based on Bayes‘ theorem with an
assumption of independence among the predictors [27]. NB
Fig. 5. Collection of 8 pituitary brain tumor MRI images from Dataset-B classifier assumes that the presence of a particular feature in a
class is unrelated to the presence of any other feature.
Fig.6 shows a collection of eight images with no tumor. Decision Tree (DT) is a supervised learning model which is
used to solve binary classification problems. DTs learn from
simple decision rules inferred from the data features and
predict the value of a target variable [28]. Random Forest is an
ensemble ML algorithm which builds multiple DTs and then
merges those together [29]. Hence, it produces results which
are more accurate. In case of DTs, if the dataset is too large
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 4

there are chances of over-fitting. So, Random Forest is used to image in the dataset has been converted into a dimension of
avoid over-fitting of data. Random Forest can be used to solve 200*200 pixels.
both classification and regression problems. Stochastic
Gradient Descent (SGD) is an efficient optimization algorithm
which minimizes the cost function by altering the values of the
parameters or coefficients of a function [30]. SGD Classifier
implements a SGD learning routine to support various loss
functions to perform classification tasks. Extreme Gradient
Boosting, XGBoost is a member of the family of boosting
algorithms. It is an efficient implementation of the Gradient
Boosted Trees algorithm which is a supervised learning
method. It is an ensemble ML technique and uses Gradient
Boosting framework for prediction [31]. Boosting is an
ensemble learning technique. It combines predictors with low Fig. 8. Image pre-processing (Dataset-A)
accuracy and converts them into a model with an improved
accuracy [32]. In gradient boosting the errors made by the 2) Splitting the dataset
predecessors is corrected by the predictor itself resulting in a The entire dataset has been split into training and testing
strong model with high accuracy. data with a test size of 25%.
3) ML models used
V. METHODOLOGY The following ML algorithms have been implemented to
perform the binary classification task:
The proposed methodology has been illustrated in Fig.7.
 SVM
 Logistic Regression
 KNN
 NB
 DT
 Random Forest
 SGD classifier
 XGBoost
 Gradient Boosting classifier
B. Methodology (multi-class classification)
The methodology for multi-class classification of the brain
MRI images in Dataset-B has been described in this section.

1) Data pre-processing
Data labeling
The images with no tumor have been labeled as ‗0‘, images
of glioma brain tumor as ‗1‘, images of meningioma as ‗2‘ and
pituitary tumor as ‗3‘.
Fig. 7. Flowchart of the proposed workflow Image pre-processing
A. Methodology (binary classification) Each and every image in Dataset-B has been resized into a
dimension of 200*200 pixels. For instance, the original image
In this section, the methodology that has been used to predict
shown in Fig.9 has a dimension of 350*350 pixels. It has been
whether a patient has brain tumor or not from the brain MRI
converted into a dimension of 200*200 pixels.
images in Dataset-A has been described.

1) Data pre-processing
Data labeling
The images of brain tumor have been labeled as ‗1‘ and the
images with no brain tumor as ‗0‘.
Image pre-processing
The images have been read in the gray scale (2D). To build a
classifier using ML algorithms all the images have been
converted into the same dimension. So, each image has been
resized into 200*200 pixels.
For instance, the original image as shown in Fig. 8 has a
dimension of 630*630 pixels. Its dimension has been
transformed into 200*200 pixels. Similarly, each and every Fig. 9. Image pre-processing (Dataset-B)
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 5

2) Splitting the dataset Misclassification= (21+7)/ (119+222+21+7) = 0.076

75% of the dataset has been used for training and the Therefore, it can be concluded that Gradient Boosting has
remaining 25% has been used for testing purposes. exhibited an accuracy of 92.4% and misclassification of 7.6%.
3) ML models used To get a clear picture, the evaluation metrics of all the ML
The following ML algorithms have been implemented to algorithms have been computed and tabulated. The
perform the multi-class classification task: performance comparative analysis of the different ML
 SVM algorithms based on accuracy, precision, recall and F1-score
 KNN have been depicted in Table II.
 Random Forest Accuracy= (TP+TN)/ (TP+FP+FN+TN)
 XGBoost Recall= TP/ (TP+FN)
Precision= TP/ (TP+FP)
VI. EXPERIMENTAL RESULTS AND DISCUSSION F1-score= (2*Precision*Recall)/ (Precision + Recall)
A. Binary classification results Table II
Comparison of the different ML algorithms based on
A comparative analysis of the nine algorithms has been done
accuracy, recall, precision and F1-score
based on the following performance metrics:
 Accuracy
model accuracy recall precision F1-score
 Recall
SVM 0.851 0.754 0.798 0.775
 Precision
Logistic 0.846 0.730 0.800 0.763
 F1- score Regression
 Area Under Curve –Receiver Operating KNN 0.845 0.804 0.756 0.779
Characteristics (AUC-ROC) NB 0.769 0.627 0.675 0.650
 Area Under Curve –Precision Recall (AUC-PR) DT 0.865 0.825 0.788 0.806
Random 0.908 0.952 0.811 0.876
For the evaluation of the accuracy, recall, precision and F1
Forest
score, the following 4 attributes have been used in the
SGD 0.840 0.709 0.807 0.755
measurement:
XGBoost 0.894 0.888 0.818 0.852
 True Positive (TP)
Gradient 0.924 0.944 0.850 0.895
 True Negative (TN)
Boosting
 False Positive (FP)
 False Negative (FN) From the test results shown in Table 2, it can be inferred that
Gradient Boosting classifier has exhibited the highest accuracy
The attributes for determining the performance metrics for among all the other ML models that have been implemented.
each of the 9 ML algorithms has been demonstrated in Table I. Gradient Boosting has produced an accuracy of 0.924, recall
of 0.944, precision of 0.850 and F1-score of 0.895. Random
Table I Forest follows Gradient Boosting with an accuracy of 0.908,
Attributes for the calculation of performance metrics of the recall of 0.952, precision and F1-score of 0.811 and 0.876
proposed ML algorithms respectively. After Gradient Boosting and Random Forest,
XGBoost classifier has produced the third highest accuracy of
Algorithm TP TN FP FN 0.894. It has a recall, precision and F1-score of 0.888, 0.818
SVM 95 219 24 31 and 0.852 respectively. DT classifier has an accuracy, recall,
Logistic 92 220 23 34 precision and F1-score of 0.865, 0.825, 0.788 and 0.806
Regression respectively. SVM has exhibited an accuracy of 0.851, recall
KNN 102 210 33 24 of 0.754, precision of 0.798 and F1-score of 0.775. Logistic
NB 79 205 38 47 Regression has generated an accuracy of 0.846 and recall,
DT 104 215 28 22 precision and F1-score of 0.730, 0.800 and 0.763 respectively.
Random 120 215 25 6 KNN has an accuracy of 0.845 which is less than that of
Forest Logistic Regression by only 0.001. It has produced recall,
SGD 88 222 21 38 precision and F1-score of 0.804, 0.765 and 0.779 respectively.
XGBoost 112 218 25 14 SGD classifier follows KNN with an accuracy of 0.840, recall
Gradient 119 222 21 7 of 0.709, precision of 0.807 and F1-score of 0.755. NB
Boosting classifier has been the least accurate with an accuracy of
0.769. It has produced a recall, precision and F1-score of
From the values of the attributed described in Table 1, it can 0.627, 0.675 and 0.650 respectively. Thus, it can be said that
be observed that Gradient Boosting classifier has detected 119 the performances of all the classification algorithms have been
data as true positives and 222 true negatives accurately. It has satisfactory with Gradient Boosting outperforming all the
misclassified only (21+7=) 28 data. other classifiers.
Accuracy= (119+222)/ (119+222+21+7) = 0.924
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 6

To enable a clear interpretation of the binary classification

models, ROC curves and PR curves have also been plotted.
This has been described in the following sections.

ROC curves of the different algorithms for performance

comparison

The ROC curves corresponding to SVM, Logistic

Regression, KNN, NB, DT classifier, Random Forest
classifier, SGD classifier, XGBoost classifier and Gradient
Boosting classifier have been shown in Fig. 10, 11, 12, 13, 14,
15, 16, 17 and 18 respectively.

Fig. 12. ROC curve of KNN classifier

Fig. 10. ROC curve of SVM classifier

Fig. 13. ROC curve of NB classifier

Fig. 11. ROC curve of Logistic Regression classifier

Fig. 14. ROC curve of DT classifier

> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 7

Fig. 15. ROC curve of Random Forest classifier Fig. 18. ROC curve of Gradient Boosting classifier

The AUC-ROC scores of all the ML classification

algorithms have been depicted in Table III to enable a
performance comparative analysis. The AUC-ROC curve is an
important evaluation metric for binary classification problems.
The area under the curve is a measure of the ability of a
classifier to distinguish between the two classes. An excellent
model should have an AUC near to 1 which is an indication
that the model has a good measure of separability.

Table III
Performance comparison of the models based on AUC-ROC

model AUC-ROC
SVM 0.888
Logistic Regression 0.886
Fig. 16. ROC curve of SGD classifier
KNN 0.920
NB 0.755
DT 0.847
Random Forest 0.975
SGD 0.806
XGBoost 0.973
Gradient Boosting 0.972

From the test results in Table III, it can be observed that the
AUC-ROC scores of the Random Forest, XGBoost and
Gradient Boosting classifiers are extremely close. While
Random Forest has an AUC-ROC score of 0.975, XGBoost
has an AUC-ROC of 0.973 and Gradient Boosting has an
AUC-ROC of 0.972. KNN follows Random Forest, XGBoost
and Gradient Boosting classifiers with an AUC-ROC score of
0.920. The AUC-ROC scores of SVM and Logistic
Fig. 17. ROC curve of XGBoost classifier Regression have also been quite promising. SVM and Logistic
Regression have AUC-ROC scores of 0.888 and 0.886
respectively. DT has an AUC-ROC score of 0.847 and AUC-
ROC score for SGD classifier is 0.806. NB has the least AUC-
ROC score of 0.755. Thus, it can be observed that the
performances of Random Forest, XGBoost and Gradient
Boosting classifiers stand out, each having AUC-ROC scores
very near to 1. While Gradient Boosting classifier has the
highest accuracy among the classifiers followed by Random
Forest and XGBoost, Random Forest and XGBoost both have
slightly higher AUC-ROC scores than Gradient Boosting.
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 8

AUC-ROC score of Random Forest exceeds that of Gradient

Boosting by 0.003. The AUC-ROC score of Gradient
Boosting is less than that of XGBoost by 0.001. Therefore, in
order to conclude which classifier is the most accurate in
detecting the presence or absence of brain tumor, the AUC-PR
scores of all the algorithms have been computed. This has
been described in the following section.

PR curves for performance comparison among the ML

algorithms

The PR curves corresponding to SVM, Logistic Regression,

KNN, NB, DT classifier, Random Forest classifier, SGD
classifier, XGBoost classifier and Gradient Boosting classifier
have been shown in Fig. 19, 20, 21, 22, 23, 24, 25, 26 and 27
respectively.
Fig. 21. PR curve of KNN classifier

Fig. 19. PR curve of SVM classifier

Fig. 22. PR curve of NB classifier

Fig. 20. PR curve of Logistic Regression classifier

Fig. 23. PR curve of DT classifier
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 9

Fig. 24. PR curve of Random Forest classifier Fig. 27. PR curve of Gradient Boosting classifier

Similar to the ROC curve, the PR curve is used for

evaluating the performance of the binary classification
algorithms. The AUC-PR is constructed by plotting the
precision against the recall for a single classifier at various
threshold values. The nearer the AUC-PR score is to 1, the
better the performance of the classifier.
The AUC-PR scores and the weighted average precision
(AP) across all the thresholds of the proposed ML algorithms
have been enlisted in Table IV.

Table IV
Performance comparison of the models based on AUC-PR
and AP

model AUC-PR AP
Fig. 25. PR curve of SGD classifier
SVM 0.829 0.896
Logistic 0.837 0.892
Regression
KNN 0.860 0.946
NB 0.723 0.790
DT 0.832 0.871
Random Forest 0.946 0.988
SGD 0.804 0.837
XGBoost 0.940 0.988
Gradient Boosting 0.914 0.987

From the test results demonstrated in Table IV, it can be

observed that Random Forest has the highest AUC-PR score
of 0.946 followed by XGBoost with an AUC-PR score of
0.940 and Gradient Boosting with an AUC-PR score of 0.914.
A good classifier should maintain both high precision and high
Fig. 26. PR curve of XGBoost classifier recall across the graph. Thus, it can be inferred performances
of Random Forest, XGBoost and Gradient Boosting classifiers
have been noteworthy. SVM, Logistic Regression, KNN, NB,
DT and SGD have AUC-PR scores of 0.829, 0.837, 0.860,
0.723, 0.832 and 0.804 respectively. While Random Forest has
the highest AUC-PR score NB classifier has the lowest.
Therefore, it is evident that the performances of Random
Forest, XGBoost and Gradient Boosting classification
algorithms have been quite encouraging. The evaluation
metric values of these three classifiers have been tabulated in
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 10

Table V in order to conclude which of these classifiers has SGD 2.629 0.061
exhibited the best overall performance. XGBoost 145.068 0.65
Gradient Boosting 1180.088 0.172
Table V
Evaluation metrics of Random Forest, XGBoost and As demonstrated in Table VI, NB classifier takes the least
Gradient Boosting classifiers training time among all the other algorithms. It takes a training
time of 2.362s. Gradient Boosting takes the maximum time for
metric Random XGBoost Gradient training. The training time for Gradient Boosting classifier is
Forest Boosting 1180.088s. Its prediction time is however 0.172s. The
Accuracy 0.908 0.894 0.924 prediction time for KNN is the highest. It takes 47.655s for
Recall 0.952 0.888 0.944 predicting the outcome. Logistic Regression with a prediction
Precision 0.811 0.818 0.850 time of 0.05s is the fastest among the ML algorithms in
F1-score 0.876 0.852 0.895 predicting the outcome.
AUC-ROC 0.975 0.973 0.972 The test results of the multi-class classification problem
AUC-PR 0.946 0.940 0.914 have been described in the following sections.
AP 0.988 0.988 0.987
B. Multi-class classification results
As depicted in Table V, Gradient Boosting classifier has an
accuracy of 0.924 which is higher than that of Random Forest A comparative analysis of the 4 ML algorithms has been
classifier by (0.924-0.908=) 0.016 and exceeds the accuracy of done based on the following evaluation metrics:
XGBoost classifier by (0.924-0.894=) 0.030. The precision of  Accuracy
Gradient Boosting is 0.850. Precision of XGBoost classifier is  Recall (weighted average)
0.818 which is less than that of Gradient Boosting by (0.850-  Precision (weighted average)
0.818=) 0.032. The precision of Random Forest is 0.811 which  F1-score (weighted average)
falls short of (0.850-0.811=) 0.039 from the precision of  AUC-ROC
Gradient Boosting classifier. The F1-score of Gradient
Boosting classifier is 0.895 which is higher than that of The performance comparison of the proposed algorithms on
XGBoost and Random Forest. XGBoost and Random Forest the basis of accuracy, recall, precision and F1-score has been
have F1-scores of 0.852 and 0.876 respectively. However, the demonstrated in Table VII.
AUC-ROC scores and AUC-PR scores of both Random Forest
and XGBoost are higher than that of Gradient Boosting. Table VII
Random Forest has an AUC-ROC score of 0.975 which is Performance comparison of the proposed algorithms
higher than that of Gradient Boosting by (0.975-0.972=)
0.003. The AUC-ROC score of XGBoost is higher than that of model SVM KNN Random XGBoost
Gradient Boosting by (0.973-0.972=) 0.001 only. Gradient Forest
Boosting has an AUC-PR score of 0.914 which is less than Accuracy 0.85 0.77 0.89 0.90
that of Random Forest by (0.946-0.914=) 0.032 and that of Recall(Weighted 0.85 0.77 0.89 0.90
XGBoost by (0.940-0.914=) 0.026. Therefore, after comparing Average )
all the evaluation metrics it can be concluded that Gradient Precision(Weighted 0.87 0.82 0.89 0.90
Boosting classifier has exhibited the best performance Average)
altogether with an accuracy, recall, precision, F1-score, AUC- F1-score(Weighted 0.85 0.78 0.89 0.90
ROC and AUC-PR scores of 0.924, 0.944, 0.850, 0.895, 0.972 Average)
and 0.914 respectively.
Also, the training time and prediction time of all the ML As depicted in Table VII, XGBoost has outperformed the
algorithms have been evaluated. This has been described in other models in terms of accuracy, recall, precision and F1-
Table VI. score. XGBoost has produced an accuracy, recall, precision
and F1-score of 0.90 respectively.
Table VI In order to visualize the performance of the multi-class
Comparison of training time and prediction time among the classifiers, the AUC-ROC curves of the four ML algorithms
ML algorithms have also been plotted.

model training time (s) prediction time (s) ROC curves of the multi-class classifiers for performance
SVM 39.325 12.24 comparison
Logistic 12.523 0.05
Regression The ROC curves corresponding to SVM, KNN, Random
KNN 13.079 47.655 Forest and XGBoost classifiers have been shown in Fig. 28,
NB 2.362 1.393 29, 30 and 31 respectively.
DT 24.41 0.101
Random Forest 13.268 0.172
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 11

Fig. 28. ROC curve of SVM multi-class classifier Fig. 31. ROC curve of XGBoost classifier

The AUC-ROC scores of the multi-class classification

algorithms have been tabulated in Table VIII.

Table VIII
AUC-ROC scores of the proposed models

model AUC-ROC
SVM 0.931
KNN 0.899
Random Forest 0.989
XGBoost 0.990

From the test results shown in Table VIII, it can be observed

that the AUC-ROC score of XGBoost is the highest. XGBoost
has exhibited an AUC-ROC score of 0.990. Random Forest
Fig. 29. ROC curve of KNN multi-class classifier has produced an AUC-ROC score of 0.989, followed by SVM
and KNN with AUC-ROC scores of 0.931 and 0.899
respectively.
Hence, after comparing the evaluation metrics of all the four
ML algorithms it can be concluded that XGBoost classifier
has outperformed the other models with accuracy, precision,
recall, F1-score and AUC-ROC of 0.90, 0.90, 0.90, 0.90 and
0.99 respectively. Therefore, XGBoost is the best model to
accomplish the task of multi-class classification of the images
from Dataset-B.

VII. CONCLUSION AND SCOPE OF FUTURE WORK

This paper presents a detailed overview of how ML algorithms
can be used for medical image processing. ML has improved
and paved the way for efficient diagnosis, recognition and
prediction in numerous domains of healthcare, brain tumor
detection and classification being one of them. Nine ML
Fig. 30. ROC curve of Random Forest multi-class classifier algorithms have been used to predict whether a patient has a
brain tumor or not based on a dataset comprising of brain MRI
images. The ML algorithms that have been used are Support
Vector Machine (SVM), Logistic Regression, K-Nearest
Neighbor (KNN), Naïve Bayes (NB), Decision Tree (DT)
classifier, Random Forest classifier, XGBoost classifier,
Stochastic Gradient Descent (SGD) classifier and Gradient
Boosting classifier. A performance comparison of the different
ML algorithms has been conducted based on a few
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 12

performance metrics such as accuracy, recall, precision, F1- [14] Ankit Ghosh, Purbita Kole and Alok Kole, ―Automatic Identification of
Score, AUC-ROC curves and AUC-PR curves. After the Covid-19 from Chest X-ray Images using Enhanced Machine Learning
evaluation of the test scores, it has been concluded that Techniques‖, International Research Journal of Engineering and
Technology (IRJET),vol.8, issue.9, no.115, pp.765-772, 2021.
Gradient Boosting is the best classifier among all the other
[15] Miles N. Wernick, Yongyi Yang, Jovan G. Brankov, Grigori Yourganov
ML classifiers that have been used. Also, multi-class
and Stephen C. Strother, ―Machine Learning in Medical Imaging‖, IEEE
classification has been performed on a different dataset Signal Processing Magazine, vol.27, issue.4, pp.25-38, 2010.
comprising of brain MRI images of glioma, meningioma, [16] Sanjay Saxena, Neeraj Sharma and Shiru Sharma, ―Image Processing
pituitary and no tumor using SVM, KNN, Random Forest and Tasks using Parallel Computing in Multi core Architecture and its
XGBoost classifier. The ML algorithms have been compared Applications in Medical Imaging‖, International Journal of Advanced
based on accuracy, recall, precision, F1-score, AUC-ROC Research in Computer and Communication Engineering, vol.2, issue.4,
score and it has been observed that XGBoost classifier has pp.1896-1900, 2013.
exhibited the best results. In future, one of the most important [17] Chetanpal Singh, ―Medical Imaging using Deep Learning Models‖,
improvements that can be made is adjusting the architecture so European Journal of Engineering and Technology Research, vol.6,
that it can be used during brain surgery, for classifying and issue.5, pp.156-167, 2021.
[18] Neelum Noreen, Sellappan Palaniappan, Abdul Qayyum, Iftikhar
accurately locating the tumor. Detecting the tumors in the
Ahmad, Muhammad Imran and Muhammad Shoaib, ―A Deep Learning
operating theatre can be performed in real-time conditions;
Model Based on Concatenation Approach for the Diagnosis of Brain
thus, in that case, the improvement would also involve Tumor‖, IEEE Access, vol.8, pp. 55135 – 55144, 2020.
adapting the network architecture to a 3D system. By keeping [19] Janki Naik and Sagar Patel, ―Tumor Detection and Classification using
the network architecture simple, detection in real time can be Decision Tree in Brain MRI‖, IJCSNS International Journal of
made possible. Computer Science and Network Security, vol.14, no.6, pp.87-91, 2014.
[20] Gopal S. Tandel, Antonella Balestrieri, Tanay Jujaray, Narender N.
REFERENCES Khanna, Luca Saba and Jasjit S. Suri, ―Multiclass magnetic resonance
imaging brain tumor classification using artificial intelligence
[1] Lars Kunze, Nick Hawes, Tom Duckett, Marc Hanheide and Tomáš paradigm‖, Computers in Biology and Medicine, vol.122, pp.103804-
Krajník, ―Artificial Intelligence for Long-Term Robot Autonomy: A 103860, 2020.
Survey‖, IEEE Robotics and Automation Letters, vol.3, issue.4, pp.4023- [21] Ahmad M. Sarhan, ―Detection and Classification of Brain Tumor in
4030, 2018. MRI Images Using Wavelet Transform and Convolutional Neural
[2] Li Deng, ―Artificial Intelligence in the Rising Wave of Deep Learning: Network‖, Journal of Advances in Medicine and Medical
The Historical Path and Future Outlook [Perspectives]‖, IEEE Signal Research,vol.32, issue.12, pp.15-16, 2020.
Processing Magazine, vol.35, issue.1, pp.180-187, 2018. [22] Heba Mohsen, El-Sayed A. El-Dahshan, El-Sayed M. El-Horbaty and
[3] Alok Kole, ―Design and Stability Analysis of Adaptive Fuzzy Feedback Abdel-Badeeh M. Salem, ―Classification using deep learning neural
Controller for Nonlinear Systems by Takagi-Sugeno Model based networks for brain tumors‖, Future Computing and Informatics Journal,
Adaptation Scheme‖, International Journal of Soft Computing, vol.19, vol.3, issue.1, pp.68-71, 2018.
issue.6, pp.1747-1763, 2015. [23] Arshia Rehman, Saeeda Naz, Muhammad Imran Razzak, Faiza Akram
[4] Ruimin Ke, Yifan Zhuang, Ziyuan Pu and Yinhai Wang, ―A Smart, and Muhammad Imran, ―A Deep Learning-Based Framework for
Efficient, and Reliable Parking Surveillance System With Edge Automatic Brain Tumors Classification Using Transfer Learning‖,
Artificial Intelligence on IoT Devices‖, IEEE Transactions on Intelligent Circuits, Systems, and Signal Processing, Springer, vol.39, pp.757–775,
Transportation Systems, vol.22, issue.8, pp. 4962 – 4974, 2021. 2020.
[5] P. P. Bhattacharya, Alok Kole, Tanmay Maity and Ananya [24] Hyun-Chul Kim, Shaoning Pang, Hong-Mo Je, Daijin Kim and
Sarkar,‘Neural Network Based Energy Efficiency Enhancement in
Sung Yang Bang, ―Constructing support vector machine ensemble‖,
Wireless Sensor Networks‘, International Journal of Applied
Engineering Research, vol. 9, no.22, pp. 11807-11818, 2014. Pattern Recognition, vol.36, issue.12, pp. 2757-2767, 2003.
[6] Huimin Lu, Yujie Li, Min Chen, Hyoungseop Kim and Seiichi [25] Sandro Sperandei, ―Understanding logistic regression analysis‖,
Serikawa, ―Brain Intelligence: Go beyond Artificial Intelligence‖, Biochemia Medica, vol. 24, no. 1, pp.12-18, 2014.
Mobile Networks and Applications, vol.23, pp.368-375, 2018. [26] Shichao Zhang, Xuelong Li, Ming Zong, Xiaofeng Zhu and Ruili Wang,
[7] Sanjeevani Bhardwaj and Alok Kole, ‗Review and Study of Internet of ―Efficient kNN Classification With Different Numbers of Nearest
Things: It‘s the Future‘, in Proc. IEEE International Conference on Neighbors‖, IEEE Transactions on Neural Networks and Learning
Intelligent Control, Power and Instrumentation (ICICPI-2016), Kolkata,
Systems, vol.29, issue.5, pp. 1774 – 1785, 2018.
India, 2016, pp.47-50.
[8] Daniel E. O'Leary, ―Artificial Intelligence and Big Data‖, IEEE [27] Saurabh Mukherjee Dr., Neelam Sharma, ―Intrusion Detection using
Intelligent Systems, vol.28, issue.2, pp.96-99, 2013. Naive Bayes Classifier with Feature Reduction‖, Procedia Technology,
[9] Jiaying Liu, Xiangjie Kong, Feng Xia, Xiaomei Bai, Lei Wang, Qing vol.4, pp.119-128, 2012.
Qing and Ivan Lee, ―Artificial Intelligence in the 21st Century‖, IEEE [28] S.R. Safavian and D. Landgrebe, ―A survey of decision tree classifier
Access, vol.6, pp. 34403 – 34421, 2018. methodology‖, IEEE Transactions on Systems, Man, and Cybernetics,
[10] Chinmaya Kumar Pradhan, Shariar Rahaman, Md. Abdul Alim Sheikh, vol.21, issue.3, pp. 660 – 674, 1991.
Alok Kole and Tanmoy Maity, ‗EEG Signal Analysis Using Different [29] Angshuman Paul, Dipti Prasad Mukherjee, Prasun Das, Abhinandan
Clustering Techniques‘, in Proc. International Conference on Emerging Gangopadhyay, Appa Rao Chintha and Saurabh Kundu, ―Improved
Technologies in Data Mining and Information Security, Kolkata, West Random Forest for Classification‖, IEEE Transactions on Image
Bengal, 2018, pp.99-105. Processing, vol.27, issue.8, pp.4012 – 4024, 2018.
[11] Alejandro F. Frangi, Sotirios A. Tsaftaris and Jerry L. Prince, [30] N. Deepa, B. Prabadevi, Praveen Kumar Maddikunta, Thippa Reddy
―Simulation and Synthesis in Medical Imaging‖, IEEE Transactions on Gadekallu, Thar Baker, M. Ajmal Khan and Usman Tariq, ―An AI-based
Medical Imaging, vol.37, issue.3, pp. 673 – 679, 2018. intelligent system for healthcare analysis using Ridge-Adaline Stochastic
[12] Norio Nakata, ―Recent technical development of artificial intelligence Gradient Descent Classifier‖, The Journal of Supercomputing,
for diagnostic medical imaging‖, Japanese Journal of Radiology, vol.37, vol.77, pp.1998–2017, 2021.
pp.103-108, 2019. [31] Shenglong Li and Xiaojing Zhang, ―Research on orthopedic auxiliary
[13] Subhamoy Mandal, Aaron B. Greenblatt and Jingzhi An, ―Imaging classification and prediction model based on XGBoost algorithm‖,
Intelligence: AI Is Transforming Medical Imaging Across the Imaging Neural Computing and Applications, vol.32, pp.1971–1979, 2020.
Spectrum‖, IEEE Pulse, vol.9, Issue.5, pp. 16 – 24, 2018.
> REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 13

[32] Djavan De Clercq, ZongguoWen and FanFei, ―Determinants of

efficiency in anaerobic bio-waste co-digestion facilities: A data
envelopment analysis and gradient boosting approach‖, Applied Energy,
vol.253, pp. 113570- 113572, 2019.