Volume 8, Issue 5, May – 2023 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Investigation of Lung Cancer Prediction and

Classification using CT-Scan Images by Employing
Machine Learning & Population based Techniques
D. Kalaivani1 Dr.G.Dheepa2
Research Scholar, Associate Professor,
Department of Computer Science, PKR Arts College for Department of Computer Science, PKR Arts College for
Women, Tamilnadu, India Women, Tamilnadu, India

Abstract:- According to the estimated reports of World Keywords:- Lung Cancer Prediction, Classification,
Health Organization, with over 2.6 million new cases Machine Learning, Deep Learning, Feature Selection, Data
captured and diagnosed each year, lung cancer is the Mining, Image Processing.
most prevalent cause of cancer-related deaths
worldwide. Early detection and classification of LC is I. INTRODUCTION
needed for effective analysis & treatments for better
patient outcomes. Lung cancer prediction and Lung cancer is a one of the significant public health
classification at an early stage have shown significant issue and one of the main causes of cancer-related fatalities
potential for advanced ML algorithms, particularly DL globally. Lung cancer accounts for 11.9% of all cancer
models. Early detection of lung cancer facilitates diagnoses and 18.4% of cancer deaths worldwide, according
patients to undergo timely and effective treatment, to the World Health Organization. Early detection and
considerably improving their chances of survival. The therapy [1] is essential for increasing the survival rates of
purpose of this research is to put forward an ISBSSA lung cancer patients. In predicting the presence or absence
(Improved Selection Based Squirrel Search Algorithm)- of lung cancer, machine learning and deep learning models
based machine learning approach for LC prediction and using advanced computational approaches have showed
classification employing CT-SCAN illustrations. The considerable potential. Machine learning techniques such as
suggested method makes use of a deep learning model Support Vector Machines (SVM), Random Forest, K-
called ISBSSA that has been trained on a substantial Nearest Neighbours (KNN), and Naive Bayes have been
dataset of computed tomography (CT) images in order employed with great accuracy for lung cancer prediction.
to accurately identify and classify lung cancer cells. For Deep learning models, such as Convolutional Neural
the experimental study, a Large-Scale CT and PET/CT Networks (CNN), have also been utilised successfully for
Dataset for Lung Cancer Diagnosis took from Cancer lung cancer prediction.
Imaging Archive (CIA) serves as the data source. The
LC-CIA dataset which includes CT and PET-CT One of the main challenges in lung cancer prediction
DICOM pictures of lung cancer patients as well as is the complexity and variability of the disease. Non-small
individuals who are healthy. The model is trained using cell lung cancer (NSCLC) and small cell lung cancer
appropriate machine learning algorithms along with (SCLC) are the two primary kinds of lung cancer [2].
ISBSSA such Naive Bayes Algorithm (NBA), NSCLC is the most prevalent kind of lung cancer,
Convolutional Neural Networks (CNNs), Support Vector accounting for around 88% of all occurrences. SCLC is less
Machines (SVMs), K-Nearest Neighbour (KNN) and frequent, but it is more aggressive and can spread to other
Random Forests (RFs), to predict the presence and type regions of the body quickly. To properly forecast the
of lung cancer cells in the CT & PET-CT DICOM existence of lung cancer, particular traits or biomarkers must
images which was extracted. The findings of this study be identified that can discriminate between the two kinds
show that the proposed approach is successful in and reliably predict the presence of cancer [3].
effectively predicting and classifying lung cancer cells in
CT scans, which might have significant implications for To discover these biomarkers and predict the existence
the early detection and treatment of the disease. The of lung cancer, machine learning and deep learning
comprehensive results show that 94.02% accuracy, algorithms may be utilised for feature extraction and
91.80% sensitivity, 92.76% specificity, 96% precision, selection [4-5]. Feature extraction is the process of
92% recall, 0.90 True Positive, 0.87 True Negative, and identifying significant features or qualities in raw data,
96.13% F-Score are achieved to detect and classify the whereas feature selection is the process of picking the most
HD in an effective manner, which is the advantage of critical features for effective prediction. For example, in
employing ML and DL approaches. lung cancer prediction using CNN, the CT scans of the
lungs are used as input [5], and the deep learning model
learns to extract relevant features and classify the image as
cancerous or non-cancerous. In SVM, the algorithm

IJISRT23MAY2611 www.ijisrt.com 3091

Volume 8, Issue 5, May – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
separates the data into two classes based on a hyper plane Nanglia et al. (2021) present a hybrid lung cancer
that maximally separates the data points. SVM features can classification system that combines support vector machines
be chosen using approaches like Recursive Feature (SVM) with neural networks. Lung cancer is one of the
Elimination (RFE) or Principal Component Analysis (PCA). major causes of cancer-related deaths globally, according to
Similarly, Random Forest builds an ensemble of decision the authors, and early diagnosis and proper categorization of
trees, and the features are chosen according on their lung nodules can dramatically improve patient outcomes.
relevance in the classification process [6]. SVM features can However, due to their complexity and heterogeneity, lung
be chosen using approaches like Recursive Feature nodules are difficult to classify. Based on the collected
Elimination (RFE) or Principal Component Analysis (PCA). characteristics, the scientists utilized SVM and neural
Similarly, Random Forest builds an ensemble of decision networks individually to categorize the lung nodules. The
trees, and the features are chosen according on their drawback of this model is the LC classification accuracy is
relevance in the classification process [7]. Lung cancer is a not upto the mark. Yakar et al. (2021) conducted a pilot
serious health concern that requires early diagnosis and research to evaluate the application of machine learning
treatment. Machine learning and deep learning algorithms algorithms for the prediction of radiation Pneumonitis in
have shown great promise in predicting the presence or stage III lung cancer patients. The researchers gathered
absence of lung cancer. These algorithms can be used for clinical and dissymmetric data from 54 individuals with
feature extraction and selection, data preprocessing, and stage III lung cancer who had radiation treatment. They
data augmentation to identify relevant features and developed models for the prediction of radiation
accurately predict the presence of lung cancer. With further Pneumonitis using machine learning methods, notably
research and development, machine learning and deep decision trees, random forests, and support vector machines.
learning can provide an important tool for early diagnosis The disadvantage of their study is that it is constrained by
and treatment of lung cancer [8]. Various researchers are the small sample size and single-center design, and they
trying to predict the LC at early stage with the help of DL recommend that future studies confirm their findings on
and ML models. This paper also aims to classify the LC and bigger and more diversified datasets. Using a genetic
early prediction with the help of ML and DL approach and approach for feature selection, Maleki et al. (2021)
also a new approach called ISBSSA to improve the accuracy suggested a k-NN technique for predicting the prognosis of
of LC prediction and classification at early occurrence. lung cancer. To create their k-NN model, the scientists
gathered clinical and demographic information from 879
II. BACKGROUND STUDY lung cancer patients. Age, gender, smoking status, tumor
size, and histology were just a few of the 29 potential
Shimazaki et al. (2022) describe a deep learning-based features they employed a genetic algorithm to narrow down
segmentation approach for identifying lung cancer on chest to the most important ones. They then classified patients
radiographs. According to the authors, lung cancer is the into low-, intermediate-, or high-risk categories depending
largest cause of cancer deaths globally, and early on their prognosis using a k-NN method. Their model had
identification is critical for successful treatment. Low-dose good performance in differentiating between low-,
computed tomography (LDCT) is the current gold standard intermediate-, and high-risk individuals, with an accuracy of
for lung cancer screening, although it is costly and has a 77.8% and an AUC of 0.78. A hybrid strategy for lung
high false positive rate. The algorithm produced an area tumor segmentation utilizing the support vector machine
under the receiver operating characteristic curve (AUC) of (SVM) and marker-controlled watershed algorithms is
0.67, which was much higher than the average AUC of 0.77 suggested by Vijh et al. (2021). The goal of the study was to
achieved by radiologists. The algorithm's sensitivity was precisely segment the lung tumor area, a crucial step in the
0.80 and its specificity was 0.74. Kumar et al. (2022) early detection and management of lung cancer. The form,
recommend employing machine learning to predict lung texture, and statistical data that are retrieved from the
cancer using text datasets. A dataset of 2000 patient records segmented tumor patches are used to train the SVM
with demographic information, medical history, symptoms, algorithm. 42 CT scan pictures were utilized in the
and test findings was employed. The dataset was investigation, of which 21 had benign tumors and the
preprocessed by transforming it to a bag-of-words format, remaining 21 had malignant tumors. However, more
which is a typical approach for text-based machine learning validation and testing on a bigger dataset are required to
applications. In this model, early detection was not determine whether the suggested strategy is useful in
accomplished. Binson et al. (2021) use machine learning clinical settings. A pathway-based method is used by Nancy
methods to investigate the potential of electronic nose (e- Lan Guo and Ying-Wooi Wan (2020) to find a smoking-
nose) technology for forecasting pulmonary illnesses. The associated 6-gene signature that is indicative of lung cancer
authors point out those traditional diagnostic procedures for risk and survival. The scientists found gene expression
lung disorders can be intrusive and time-consuming, profiles that potentially predict lung cancer risk and survival
resulting in diagnostic and treatment delays. E-nose using microarray data from the Gene Expression Omnibus
technology is a non-invasive and quick method of detecting database. To find the gene signature, they applied a
volatile organic compounds (VOCs) in exhaled breath, pathway-based method, which groups genes according to
which can provide important information regarding how they operate in biological networks. To confirm the
pulmonary illnesses. SVM and XGBoost both performed gene signature, the scientists also employed machine
well in this model, with SVM attaining an accuracy of learning techniques including support vector machines
85.4% and XGBoost reaching an accuracy of 86.2%. (SVM) and random forests. Further study and development

IJISRT23MAY2611 www.ijisrt.com 3092

Volume 8, Issue 5, May – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
of the found 6-gene signature might result in a valuable tool
for determining lung cancer risk and developing a tailored
treatment strategy.

The purpose of this research study is to study and

introduce a ML with GA method to classify the LC disease;
make early-stage LC detections using CT scan images;
maximize the accuracy level.

III. PROPOSED METHODOLOGY

 Issbssa:
The proposed method for LC prediction is carried out
by using CNN, NBA, SVM, RF, KNN and ISBSSA. The
ML and DL models used to feature selection and extraction.
Here the ISBSSA is portrayed for the LC prediction and it is
basically inspired by the food forging behavior of the flying
squirrels in the real life. Depends on the weather condition
the flying squirrel search for food and store them for future
use. If the climate is hot the squirrels will fall down from a
tree. It will move out and rapidly search for food for daily
needs. It will eat acorns which are available in the hot
climate. Once they consumed it will search food for winter.
During bad weather the hickory nuts will help squirrels to
satisfy the needs. So the process will be continuously
depends on the weather condition in the area. Let’s take this
into mathematical model, the following hypothesis are taken
into account.

 In a forest, the flying squirrel can be counted as 1, it

will stay on the tree for whole year
 The foraging behavior of FS is dynamic depends on the
climatic condition and resource available
 Only 3 types of trees in forest hickory trees, oak trees
and normal trees.
 The ‘n’ is the number of squirrels searching for food ‘f’
in the trees ‘t’ to jump and forage. If winter s will jump
to t1 and vice versa.
 Using JSM the squirrels will jump into one dimensional
and two dimensional search spaces.
 It has ability to change the locations ‘l’ where it will be
recorded for measure.
 Position is calculated and later the best solution is
measured.
 Increment the counter and calculate the fitness value
for new solution.
Fig 1 Flow Diagram of the Proposed ISBSSA

In real time scenario, Applying the biotic and abiotic

search in image processing that helps to identify the LC in
the CT-PET scan images loaded in the system and classify
the disease and generates the output. As the ISASSB
generates the fitness value, in real time the CGM (Conjugate
Gradient Method) and ECS (Edge Colour Segmentation) is
used to identify the spot to detect the LC at early stage.

 NBA:
To examine the wide range of supervised and non
supervised learning NBA is used. Suppose in a CIA dataset
with only two features (age and smoking history) and a
binary target variable indicating whether or not the patient

IJISRT23MAY2611 www.ijisrt.com 3093

Volume 8, Issue 5, May – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
has lung cancer. We can train a Naive Bayes model on this  Output layer- OL: The final layer produces a binary
data by calculating the probability of each feature given the classification output indicating whether or not the input
class (i.e. the probability of each feature given lung cancer image shows signs of lung cancer. Once the LC is
and the probability of each feature given no lung cancer), predicted the binary values are measured and
and then combining these probabilities using Bayes' performance is calculated.
theorem. For example, suppose we calculate the following
probabilities: A loss function that calculates the difference between
the expected result and the actual output is minimized
 P(age=50|lung cancer) = 0.2 during training by adjusting the model's weights via back
 P(age=50|no lung cancer) = 0.3 propagation and gradient descent. Once the model has been
 P(smoking history=heavy|lung cancer) = 0.8 trained, it may be used to determine from the characteristics
 P(smoking history=heavy|no lung cancer) = 0.4 it extracts from an image if a fresh medical imaging has
indicators of lung cancer. CNN has shown more accuracy in
In observing a new patient with age 50 and a heavy terms of LC prediction.
smoking history, Naive Bayes' theorem is used to calculate
the probability of each class:  SVM:
SVM works on separating the classes in a HD feature
P(lung cancer | age=50, smoking history = heavy) ∝ space. In of LC prediction SVM learns the decision
P(age=50 | lung cancer) * P(smoking history = heavy | lung boundary that separates the medical CT scan images into 0
cancer) * P(lung cancer) = 0.2 * 0.8 * P(lung cancer) and 1 that indicates disease and non disease. Model training
and model evaluation steps will be carried out after FE and
P(no lung cancer | age=50, smoking history = heavy) FS.
∝ P(age=50 | no lung cancer) * P(smoking history = heavy |
no lung cancer) * P(no lung cancer) = 0.3 * 0.4 * P(no lung  Input:
cancer) CIA dataset medical images {x1, x2, xn}, where each
xi is a vector of extracted features, and corresponding labels
Compare these probabilities to make a LC prediction. {y1, y2, yn} indicating whether the image shows signs of
If the probability of lung cancer is higher than the lung cancer (1) or not (0).
probability of no lung cancer, we predict that the patient has
lung cancer. Otherwise, we predict that the patient does not  Output:
have lung cancer. In NBAs the prediction level and A trained SVM model that predicts the presence or
accuracy is shown in the findings. absence of lung cancer in new medical images.

 CNN:  Split the CIA data into training and test sets.
CNNs are used to analyze medical images loaded in  Select a subset of features from the training data that
the dataset (CIA Images), such as chest X-rays or CT scans, are most relevant for lung cancer prediction, using a
to detect signs of lung cancer in all stages. After feature selection technique such as mutual information
preprocessing, feature selection and extraction the system or L1 regularization.
start functioning to identify the prediction levels. Here are  Train an SVM on the selected features using the
the LC prediction steps when implementing through CNNs, training set, optimizing the hyperparameters using
cross-validation.
 Input layer: The input to the model is a CIA dataset  Evaluate the performance of the trained SVM on the
medical image of the lungs, such as a PET/DICOM and test set, using PEM shown below.
CT scan images.  Adjust the hyperparameters or feature selection criteria
 Convolutional layer: The 1st layer applies a set of and repeat steps 3 and 4 until an acceptable
filters to the input image to extract features that are performance is reached if the SVM's performance is not
relevant for detecting LC. adequate.
 ReLUAF activation: A rectified linear unit (ReLUAF)  By removing pertinent information and using the learnt
activation function is applied to the output of the decision boundary, the model may be used to predict
convolutional layer to introduce non-linearity into the the presence or absence of lung cancer in CIA medical
CNN model. images after training.
 Pooling layer - PL: The output of the activation  The SVM shows the remarkable performance in LC
function is passed through a PL to down sample the prediction compares to NBA and CNN which is shown
image spotted in the CIA dataset. in the diagram.
 Dropout layer - DO: A dropout layer is added to
prevent overfitting by randomly dropping out a fraction  KNN:
of the units in the layer during training. In case of KNN, the following steps are followed for
 Fully connected layer - FCL: The pooled output is LC prediction, the steps are,
flattened and passed through one or more fully
connected layers, which use the extracted features to  Preprocessing
make a final prediction.  Extracting from CIA dataset

IJISRT23MAY2611 www.ijisrt.com 3094

Volume 8, Issue 5, May – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
 Splitting datasets to train and test IV. THE ISBSSA ALGORITHM
 Choose value of k-distance metric by using Euclidian
distance method  Begin : Define the input
 Train the KNN model by storing feature vectors  Random positions for ‘n’ of floating squirrels (Sq)
 Predict the class  calculate the fitness of Sq
 Evaluate the performance  Sort positions of Sq based on fitness
 Adjust the hyper parameters  Announce the floating Sq on 3 type of trees (hickory
 Find the K-Nearest Neighbour trees, acorn, nut tree)
 Elect random and float some Sq move from normal
As KNN is using the distance vector metric, the trees to hickory trees
prediction accuracy is little high compares to CNN, NBA  for t=1 to n1 (n=total Sq)
and SVM.  Evaluate the season
 Update the foraging value
 RF:  Update the fitness
By constructing several DTs (Decision Trees) and  Randomly reposition the Sq
combining their projections to increase accuracy. The RF  Update the lowest value
selects the number of trees and other hyperparameters after  The final position of h_tree considered as best solution
FS and FE: In order to use random forest, you must provide
 End
the number of trees to be built, the maximum depth of each
tree, and additional hyperparameters like the minimum
Select the features that appear the most frequently:
number of samples needed to divide a node. To boost the
Following the completion of the SSA process, the frequency
model's performance, these hyperparameters will be of each feature in the best solutions can be calculated. The
modified using methods like cross-validation.
most often occurring features can be chosen as the best
feature subset for lung cancer prediction. Use the selected
 Input:
features to train a classifier: Once the optimal feature subset
is selected, it can be used to train a classifier such as SVM
 Dataset: CIA medical dataset (preprocessed) or KNN for lung cancer prediction on the test set. Parameter
 Number of trees: n_trees tuning is essential to find the best result.
 Maximum depth of each tree: max_depth
 Minimum number of samples required to split a node:  Search Process:
Min, Samples, Split
 Initialize the best solution as the one with the highest
 Output: fitness value - HFS.
Random Forest model for lung cancer prediction  For each Sq in the population, calculate the movement
distance based on its fitness and the fitness of the best
 Split the dataset into training and test sets solution.
 For each tree i in n_trees  Update the location of the Sq by adding a random
 Randomly select a subset of the training data displacement vector to its current position.
 Construct a decision tree using the selected data with a  Assess the fitness of the new solution and compare it
maximum depth of max_depth and minimum samples with the previous one.
required to split a node of min_samples_split  If the new solution is better than the previous one,
 Store the decision tree in the ensemble update the best solution and the population.
 Return the Random Forest model
 Use the Random Forest model to predict the lung V. MATLAB R2020A IMPLEMENTATION
cancer status for new patients:
 For each patient's medical data, traverse each decision The performance evaluation of ISBSSA and other ML
tree in the ensemble models such as NBA, CNNs, SVM, KNN, and RF for LC
 Record the predicted class for each tree prediction is done using the MATLAB R2020a tool. It is
 Assign the new patient to the class with the most votes used to analyse the proposed work and showcase the
among the trees graphical representation of the model's efficacy. For
conducting the simulation, NS2.35 is used. The number of
 As RF used decision tree method, the accuracy of LC
iterations was 800, but 1000 iterations are required to reach
prediction is high and shows the evident result and
the termination condition. The CIA datasets (355 instances
shown in the results.
with 4 types of cancer support in 251,135 images) used for
the implementation process are clearly shown in Table 1.

IJISRT23MAY2611 www.ijisrt.com 3095

Volume 8, Issue 5, May – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Table 1 CIA Datasets for LC prediction (CT Images)
Image Statistics
Modalities CT/PT DICOM Images
Number of Instances 355
Number of Studies 436
Number of Series 1295
Number of PE Images 251135
Attributes - Clinical LC
1.Patient ID 2.Sex
2.Age 4.Weight
5.T-Stage 6.N-Stage
7.M-Stage 8.Histopathological Grading
9.Smoking History 10.Target Value

 Performance Metrics  Sensitivity and Specificity - Detects accurately all the

The following metrics are used in this analysis to positive and negative instances in the image given in
compare the performance of the proposed work ISBSSA •CIA dataset for ISBSSA and other ML models.
along with NBA, CNNS, SVM, KNN and RF. The results
are shown in the graphical format where the readers can VI. FINDINGS AND DISCUSSIONS
compare the analysis. The metrics are,
 Accuracy in LC prediction
 Accuracy in LC prediction - Shows the efficiency of Figure 2 shows the detailed analysis of LC prediction
ISBSSA and other ML models for LC prediction.. and its accuracy. Among all the ML and DL models,
 True Positive & True Negative - Accuracy of LC ISBSSA outperforms all other algorithms and stands at the
prediction where the system actually does. top. The ISBSSA initiates the process at the beginning of
 False Positive & False Negative - Failure rate of the the image search in the LC-CIA dataset, where the depth
system towards LC prediction. ratio is calculated and measured by its coherent value. The
 Precision - Calculates the amount of TP rates among all depth ratio and, due to optimisation of images, the accuracy
positive predictions/cases made by system. rate is high in ISBSSA compared to other models. The rate
 Recall - Calculates the amount of TP out of actual is calculated as 0-1 and ISBSSA shows outstanding
positive cases/instances in the given CIA dataset. performance of 1 in terms of LC prediction.
 Fault/Error Comparison - It shows the error percentage
of the proposed ISBSSA and other ML models.

Table 2 Accuracy Analysis of LC prediction

Accuracy in LC Prediction
Rounds NBA CNN SVM KNN RF ISBSSA
0 0 0 0 0 0 0
200 0.2 0.2 0.2 0.2 0.2 0.5
400 0.3 0.4 0.4 0.4 0.4 0.7
600 0.5 0.5 0.5 0.4 0.6 1
800 0.5 0.6 0.6 0.5 0.7 1

Fig 2 Accuracy in LC Prediction

IJISRT23MAY2611 www.ijisrt.com 3096

Volume 8, Issue 5, May – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
 True Positive and True Negative
Figure 3 shows a comparative analysis of DM and ML models in terms of TPR and TNR. As far as LC prediction is
concerned, the TPR and TNR stand top in ISBSSA whereas NBA, CNN, SVM, KNN and RF models are relatively good. As the
number of iterations increases the TPR and TNR rate is gradually increases in all the approaches. As ISBSSA is capable of
working with both binary and normalized data, the complete search space is calculated and disease spot is identified in the search
space.

Table 3 TP & TN Analysis of LC Prediction

True Positive & True Negative
Rounds NBA CNN SVM KNN RF ISBSSA
0 0 0 0 0 0 0
200 0.3 0.4 0.5 0.5 0.6 0.7
400 0.3 0.4 0.6 0.6 0.6 0.9
600 0.5 0.6 0.6 0.7 0.9 1
800 0.5 0.6 0.7 0.8 0.9 1

Fig 3 True Positive & True Negative

 False Positive and False Negative

Figure 4 portrays the detailed analysis of False Positive and False Negative of the proposed ML and DL approaches.
ISBSSA showed the evident success in minimizing FPR and FNR. On the chosen datasets with features and instances, the new
ISBSSA ML technique outperformed the other DL methods due to its high accuracy rate and measured the proportion of negative
cases that are incorrectly identified as positive and vice versa. The new model ISBSSA performs well in prediction and
classification of lung cancer due to the computational methods, and the results are given below. Minimum of 0.05 were plotted
where the number of iterations is 800 in ISBSSA. The model works out well in all iterations/rounds for all the optimized images
loaded in the datasets. The spotting of colours in CT-PET scan images is done by CGDM and ECSM method.

Table 4 FP & FN Analysis of LC Prediction

False Positive and False Negative
Rounds NBA CNN SVM KNN RF ISBSSA
0 0 0 0 0 0 0
200 0.6 0.5 0.4 0.3 0.3 0.1
400 0.5 0.4 0.4 0.3 0.3 0.1
600 0.5 0.3 0.3 0.2 0.2 0.1
800 0.4 0.3 0.3 0.2 0.2 0.05

IJISRT23MAY2611 www.ijisrt.com 3097

Volume 8, Issue 5, May – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Fig 4 False Positive & False Negative

 Precision and Recall

Figure 5 present the precision and recall analysis of NBA, CNN, SVM, KNN, RF and ISBSSA. It shows the number of
iterations and performance rate by calculating the accuracy of LC prediction at early stage. Despite the fact that the number of
rounds increases, the P&R level is high in the proposed ML and DL approach. ISBSSA reached 1 performance/accuracy level
where the other approaches are marginally low.

Table 5 Precision & Recall Analysis of LC Prediction

Precision and Recall
Rounds NBA CNN SVM KNN RF ISBSSA
0 0 0 0 0 0 0
200 0.2 0.2 0.2 0.2 0.2 0.4
400 0.3 0.3 0.4 0.4 0.4 0.6
600 0.5 0.5 0.4 0.4 0.5 0.8
800 0.5 0.6 0.6 0.5 0.8 1

Fig 5 Precision & Recall

 Fault/Error Comparison Rate

Figure 6 illustrates the fault and error comparison rate. It shows that the error rate decreases as the number of iterations
increases. Due to the optimisation of ISBSSA, the error or fault rate is low, and the approach outperforms other DL and ML
models like NBA, CNN, SVM, KNN, RF, and ISBSSA. As the LC prediction is done at an early occurrence, the CT scan image
optimisation is much needed for fast forwarding the process and getting quick results.

IJISRT23MAY2611 www.ijisrt.com 3098

Volume 8, Issue 5, May – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Table 6 Error Comparison Rate Analysis of LC Prediction
Fault/Error Comparison Rate
Rounds NBA CNN SVM KNN RF ISBSSA
0 0.8 0.7 0.7 0.6 0.6 0.3
200 0.8 0.7 0.6 0.5 0.5 0.3
400 0.7 0.7 0.5 0.4 0.4 0.2
600 0.6 0.6 0.5 0.4 0.3 0.1
800 0.4 0.5 0.5 0.4 0.3 0.1

Fig 6 Error/Fault Comparison

 Sensitivity and Specificity

The performance analysis S&S of the ISBSSA and other ML and DL models towards LC prediction is shown in Figure 7.
Figure 8 displays the evaluated outcomes. It is noted that the proposed ML technique produced remarkable performance with
enhanced results. The new SDSM is employed in ISBSSA and provides better performance than any previous ML technique. Due
to precise classification and extraction, the suggested ML techniques perform well up to maximum level 1 in terms of anticipating
the LC at an early stage.

Table 7 Sensitivity & Specificity Analysis of LC prediction

Sensitivity and Specificity
Rounds NBA CNN SVM KNN RF ISBSSA
0 0 0 0 0 0 0
200 0.2 0.2 0.2 0.4 0.5 0.7
400 0.3 0.3 0.3 0.5 0.6 0.9
600 0.4 0.4 0.4 0.5 0.7 1
800 0.5 0.55 0.5 0.5 0.7 1

Fig 7 Sensitivity and Specificity

IJISRT23MAY2611 www.ijisrt.com 3099

Volume 8, Issue 5, May – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
VII. CONCLUSION [6] Ahsan Bin Tufail, Yong-Kui Ma, Mohammed K. A.
Kaabar, Francisco Martínez, A. R. Junejo, Inam
In order to improve prediction accuracy and detect LC Ullah, Rahim Khan, "Deep Learning in Cancer
at an early stage, the research work offered a prediction and Diagnosis and Prognosis Prediction: A Minireview
classification of lung cancer utilizing multiple ML on Challenges, Recent Trends, and Future
techniques, including NBA, CNN, SVM, KNN, RF, and Directions", Computational and Mathematical
ISBSSA. The suggested method makes use of a dataset of Methods in Medicine, vol. 2021, Article ID 9025470,
CIA PT-CET CT scan images, and 355 cases with 15 or 28 pages, 2021.https://doi.org/10.1155/2021/9025470
more features are utilised to choose the set of feature values [7] Shimazaki, A., Ueda, D., Choppin, A. et al. Deep
using an ISB-SSA genetic algorithm (GA). The trained and learning-based algorithm for lung cancer detection on
test values are compared to the available datasets to make a chest radiographs using the segmentation method. Sci
prediction. 800+ iterations are carried out which contains Rep 12, 727 (2022). https://doi.org/10.1038/s41598-
patient information, such as medical history, lab results, and 021-04667-w
imaging data, to predict the likelihood of a patient having a [8] C. Anil Kumar, S. Harish, Prabha Ravi, Murthy
specific type of habits like smoking, alcohol consumption SVN, B. P. Pradeep Kumar, V. Mohanavel, Nouf M.
etc. The results show that the ML ISBSSA technique Alyami, S. Shanmuga Priya, Amare Kebede Asfaw,
prediction is more accurate and faster than the other ML, "Lung Cancer Prediction from Text Datasets Using
DL prediction models and successfully classified the LC. Machine Learning", BioMed Research International,
For the medical field, it is also a time-saving and user- vol. 2022, Article ID 6254177, 10 pages, 2022.
friendly application. It is noted that ML decision support https://doi.org/10.1155/2022/6254177
system is more suitable for LC prediction. [9] V. A. Binson, M. Subramoniam, Y. Sunny and L.
Mathew, "Prediction of Pulmonary Diseases With
The suggested method has limitations since it only Electronic Nose Using SVM and XGBoost," in IEEE
uses a set of attributes that have been set up, and prediction Sensors Journal, vol. 21, no. 18, pp. 20886-20895, 15
and accuracy levels may differ from instance to case. The Sept.15, 2021, https://doi.org/10.1109/JSEN .2021.
algorithm may be improved in the future to forecast LC with 3100390.
skin disease accuracy rates even better and to make the [10] Pankaj Nanglia, Sumit Kumar, Aparna N. Mahajan,
system more adaptable based on the stability of the patients. Paramjit Singh, Davinder Rathee, A hybrid algorithm
for lung cancer classification using SVM and Neural
REFERENCES Networks, ICT Express, Volume 7, Issue 3, 2021,
Pages 335-341. https://doi.org/10.1016/j.icte. 2020.
[1] Li, P., Wang, S., Li, T., Lu, J., HuangFu, Y., & 06.007.
Wang, D. (2020). A Large-Scale CT and PET/CT [11] Yakar M, Etiz D, Metintas M, Ak G, Celik O.
Dataset for Lung Cancer Diagnosis [Data set]. The Prediction of Radiation Pneumonitis With Machine
Cancer Imaging Archive. https://doi.org/10.7937/ Learning in Stage III Lung Cancer: A Pilot Study.
TCIA.2020.NNC2-0461 Technology in Cancer Research & Treatment. 2021;
[2] Patra, R. (2020). Prediction of Lung Cancer Using 20. https://doi.org/10.1177/15330338211016373
Machine Learning Classifier. In: Chaubey, N., [12] Negar Maleki, Yasser Zeinali, Seyed Taghi Akhavan
Parikh, S., Amin, K. (eds) Computing Science, Niaki, A k-NN method for lung cancer prognosis
Communication and Security. COMS2 2020. with the use of a genetic algorithm for feature
Communications in Computer and Information selection, Expert Systems with Applications, Vol
Science, vol 1235. Springer, Singapore. 164,2021,113981. https://doi.org/10.1016/
https://doi.org/10.1007/978-981-15-6648-6_11 j.eswa.2020. 113981.
[3] Kadir, T., & Gleeson, F. (2018). Lung cancer [13] Vijh, S., Sarma, R., & Kumar, S. (2021). Lung
prediction using machine learning and advanced Tumor Segmentation Using Marker-Controlled
imaging techniques. Translational Lung Cancer Watershed and Support Vector
Research, 7(3), 304-312. Machine. International Journal of E-Health and
https://doi.org/10.21037/tlcr.2018.05.15 Medical Communications (IJEHMC), 12(2), 51-64.
[4] Nemlander E, Rosenblad A, Abedi E, Ekman S, https://doi.org/10.4018/IJEHMC.2021030103
Hasselström J, Eriksson LE, et al. (2022) Lung [14] Nancy Lan Guo, Ying-Wooi Wan, Pathway-based
cancer prediction using machine learning on data identification of a smoking associated 6-gene
from a symptom e-questionnaire for never smokers, signature predictive of lung cancer risk and survival,
formers smokers and current smokers. PLoS ONE Artificial Intelligence in Medicine, Volume 55, Issue
17(10): e0276703. 2, 2020, Pages 97-105. https://doi.org/10.1016/j.
https://doi.org/10.1371/journal.pone.0276703 artmed. 2012. 01.001.
[5] Nithyanandh S and Jaiganesh V. (2020), [15] Nithyanandh S and Jaiganesh V, (2020), Quality of
Reconnaissance Artificial Bee Colony Routing service enabled intelligent water drop algorithm
Protocol to Detect Dynamic Link Failure in Wireless based routing protocol for dynamic link failure
Sensor Network. International Journal of Scientific detection in wireless sensor network. Indian Journal
& Technology Research, 10(10), 3244–3251. of Science and Technology. 2020; 13(16), 1641-1647.
https://doi.org/10.35940/ijstr.b2271.0986231 https://doi.org/ 10.17485/IJST/v13i16.19

IJISRT23MAY2611 www.ijisrt.com 3100

Volume 8, Issue 5, May – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
[16] Park, S., Lee, S.M., Lee, K.H. et al. Deep learning-
based detection system for multiclass lesions on chest
radiographs: comparison with observer readings.
European Radiology, 30, 1359–1368 (2020).
https://doi.org/10.1007/s00330-019-06532-x
[17] Yoo H, Kim KH, Singh R, Digumarthy SR, Kalra
MK. Validation of a Deep Learning Algorithm for
the Detection of Malignant Pulmonary Nodules in
Chest Radiographs. JAMA Netw Open. 2020;3(9):
e2017135.https://doi.org/10.1001/jamanetworkopen.2
020.17135
[18] Hwang EJ, Park S, Jin K, et al. Development and
Validation of a Deep Learning–Based Automated
Detection Algorithm for Major Thoracic Diseases on
Chest Radiographs. JAMA Netw Open. 2019;2
(3):e191095.https://doi.org/10.1001/jamanetworkope
n.2019.1095
[19] Nithyanandh S and Jaiganesh V, (2020), Dynamic
Link Failure Detection using Robust Virus Swarm
Routing Protocol in Wireless Sensor Network,
International Journal of Recent Technology and
Engineering, 8(2), 1574-1578. https://doi.org/
10.35940/ijrte.b2271.078219
[20] Eui Jin Hwang, Sunggyun Park, Kwang-Nam Jin,
Jung Im Kim, So Young Choi, Jong Hyuk Lee, Jin
Mo Goo, Jaehong Aum, Jae-Joon Yim, Chang Min
Park, Deep Learning-Based Automatic Detection
Algorithm Development and Evaluation Group,
Development and Validation of a Deep Learning–
based Automatic Detection Algorithm for Active
Pulmonary Tuberculosis on Chest Radiographs,
Clinical Infectious Diseases, Volume 69, Issue 5, 1
September 2019, Pages 739–747, https://doi.org/
10.1093/cid/ciy967
[21] Arnaud Arindra Adiyoso Setio, Alberto Traverso,
Thomas de Bel, Moira S.N. Berens, et.al, Validation,
comparison, and combination of algorithms for
automatic detection of pulmonary nodules in
computed tomography images: The LUNA16
challenge, Medical Image Analysis, Volume 42,2017,
Pages 1-13. https://doi.org/10.1016/j.media. 2017.06.
015.
[22] Dritsas, E.; Trigka, M. Lung Cancer Risk Prediction
with Machine Learning Models. Big Data Cogn.
Comput. 2022, 6, 139. https:/doi.org/10.3390/ bdcc
6040139
[23] Marjolein A. Heuvelmans, Peter M.A. van Ooijen,
Sarim Ather, Carlos Francisco Silva, Daiwei Han,
Claus Peter Heussel, William Hickes, Hans-Ulrich
Kauczor, et.al, Lung cancer prediction by Deep
Learning to identify benign lung nodules, Lung
Cancer,Volume 154,2021, Pages 1-4. https://doi.org/
10.1016/j. lungcan. 2021.01.027.

IJISRT23MAY2611 www.ijisrt.com 3101