SVM &GA-CLUSTERING BASED FEATURE SELECTION APPROACH FOR BREAST CANCER DETECTION

International Journal on Soft Computing, Artificial Intelligence and Applications (IJSCAI), Vol.9, No.4,
November 2020
DOI :10.5121/ijscai.2020.9401 1
SVM &GA-CLUSTERING BASED FEATURE
SELECTION APPROACH FOR BREAST
CANCER DETECTION
Rashmi Priya1
and Syed Wajahat Abbas Rizvi2
1
Assistant Professor GD Goenka University ,India
2
Amity University, Uttar Pradesh ,India
ABSTRACT
Mortality leading among women in developed countries is breast cancer. Breast cancer is women's second
most prominent cause of cancer mortality worldwide. In recent decades, women's high prevalence of breast
cancer has risen dramatically. This paper discussed several data analysis methods used to detect breast
cancer early. Breast cancer diagnosis distinguishes benign and malignant breast lumps. Using data
processing tools, we tackled this disease analysis. Data mining is an important step of library discovery
where intelligent methods are used to detect patterns. Several clinical breast cancer studies were
conducted using soft computing and machine learning techniques. Sometimes their algorithms are easier,
easier, or more comprehensive than others. This research is focused on genetic programming and machine
learning algorithms to reliably identify benign and malignant breast cancer. This study aimed to optimise
the testing algorithm. We used genetic programming methods to choose classification machines' best
features and parameter values. Data mining is an important step of library discovery where intelligent
methods are used to detect patterns. We are analysing data accessible from the U.C.I. deep-learning data
set in Wisconsin. In this experiment, we equate four Weka clustering strategies with genetic clustering. A
comparison of results reveals that sequential minimal optimization (S.M.O.) is better than I.B.K. and B.F.
Tree processes, i.e. 97.71%.
KEYWORDS
S.M.O., Breast cancer, Machine learning, Feature selection, and WEKA
1. INTRODUCTION
Breast cancer is the most prevalent non-skin cancer in women and the second-largest cause of
cancer mortality in women. [1]. [1]. "Mammography usually represents thick-area breast cancer
and clusters. A normal benign mass has a circular border, circumscribed and round, but malignant
cancer usually has a suspected raw and fuzzy boundary. "[2],[3].
Nowadays, demand for machine learning grows until it becomes an operation. Sadly, machine
learning also takes skills and is a field with very high barriers. The creation of a successful
machine learning model involves many skills and experience, including pre-processing steps,
feature selection, and classification processes. Using computer analysis and machine learning
methods in medical fields is prevalent, as these techniques can be regarded as a great aid in
decision-making processes for medical professionals. A vast variety of databases are being used
for breast cancer cases that are helpful in supporting scientific and academic studies and even
more in integrating previous field computer analysis and machine learning.

November 2020
2
In certain fields and implementations, solving such a problem is based on retrieving features from
the original images obtained in the physical world and organised as vectors. The processing
system 's efficiency depends heavily on the correct choice of these vectors. But, in many
situations, problem solving becomes almost impossible due to the excessive dimensionality of
these vectors or anomalies that may exist in the results. Therefore, reducing the size of the dataset
samples to a more suitable size is sometimes helpful and often necessary, even though this
reduction may lead to minor information loss.
Accurate and accurate data will be gathered to help the doctor's early diagnosis and treatment of
illness, both stable and malignant, using an exact model. This will save doctors time and improve
efficiency. This article reflects on how benign or malignant breast cancer diagnosis is, also at an
early stage. Forecast criteria are based on breast cancer symptoms. This paper's data set contains
32 attributes.
A breast cancer diagnosis may be useful in predicting the outcomes of complex diseases or
recognising the molecular nature of the tumour. Many methods for examining and identifying
trends of breast cancer. This paper compares empirically the utility of three classical tree
classifiers designed to specifically evaluate their effects. Traditional SVM preparation normally
requires Q.P. Set, and it takes longer to solve Q.P. Optimization problem, particularly for large
dataset problems. SVM planning is slow, and the creation of large data sets takes time. The
SMO-SVM method minimises data, is more reliable and easier to implement[4]. We suggest a
hybrid SVM-GA (Support vector machines with genetic attributes) approach to achieve optimal
results on Dataset breast cancer.
2. LITERATURE SURVEY
Machine learning techniques usually refer to life scientific analysis. Numerous research focused
on medical diagnostic technology has been written. These studies have applied different solutions
to the problem and achieved detailed classification precision[5] using an artificial cortical
network to assess breast cancer therapy. They have tested their system on a limited set of data,
but results suggest they understand actual survival. And al.[6] Also in breast cancer patients, a
naïve bay, decision tree, and neural backpropagation network were used.
Though the results were high (about 90% accuracy), they were not appropriate because the data
were split into two groups: one for more than five years of life and the other for those who died in
five years. Findings became meaningless. [7] Program pick approach to usability assessment of
feature selectors. This seeks a simple, coherent set of functions without losing the predictive
precision dimension. Using a ranking algorithm applies confidence to characteristics. [8]
Proposed hybrid GA / SVM approach using fuzzy logic to minimise the initial problem size.
Identify a sub-set of balanced genes, which are then checked by SVM. [9] This analysis aimed to
compare the performance of the Artificial Neural Network.
(ANN) and Vector Machine Support (SVM) for liver cancer classification. On BUPA Liver
Disease Dataset, both model accuracy, durability, precision, and performance were contrasted
and validated. Curved Field (A.U.C.). [10] used in tandem with heart disease modelling in mining
and genetic algorithms. The proposed approach used Gini genetic mutation index statistics for
interface process and crossover. They used a technique for gathering consistency functions.
3. CLASSIFICATION
Classification is one of the data mining techniques specifically used to analyse and assign a given
dataset to a particular class[11]. This method is designed to remove classification errors.

November 2020
3
Classification enables model extraction that determines the forms for a given data set. Three
different testing approaches are used in information processing: guided and unrestricted study.
[12] [12]. Classification is a first step in evaluating many situations. The overarching aim is to
enhance market comprehension or forecasts. Analysis has proposed many common ways of
classification techniques. Data mining 's central operation is designing accurate classification
systems. A vital mission for data processing and machine learning research. Like decision-
making trees, naive- Bayesian systems, serial minB.B.F.um optimization (S.M.O.), I.B.K., B.F.
Chain modes of grouping methods, etc.
4. FEATURE SELECTION
Increased use of computers from both directions contributes to comprehensive data processing.
This data are large, systematically interconnected data to identify acceptable patterns, making
data mining a crucial area for data processing, prediction, and other activities. It has joined a
complex science area to address real- time theoretical problems. Big data mining is used in many
areas where data analysis is needed. These data mining and creation techniques were commonly
used at various levels, such as pattern recognition, etc., and collecting apps plays an important
role in virtually every field.
The collection attempts to assess the smallest possible subset of characteristics. The architecture
selects the foundation of original features by removing redundant and unused dimensionality
features without losing information. Until data-mining activities are implemented, the pre-
processing step is critical. Mining accuracy, measurement time, and test comprehension are
improved. Three filtering strategies comprises of philtres, wrappers and embedded approaches.
As discussed in [16], the Filter selects the function without the classifier type used. The
advantage of this method is that it is only necessary to select features once, it is simple and
irrespective of the classifier used. This procedure, however, lacks the classifier relationship; each
feature is interpreted separately from functional dependence. Wrapper's approach depends on
classification. Classifier results are used to assess the goodness of the specified feature or
attribute. Another method has the benefit that the filtering cycle eliminates the downside, which
is simpler than the filtering system as it still takes all the dependencies. The next embedded
approach is to combine a philtre algorithm with a wrapper approach to find an optimal sub-set in
the classification structure. This method 's advantage is less expensive and less vulnerable than
wrapper approach. Different feature selection applications over the past two decades:
• Text mining
• Image processing and computer vision
• Industrial applications
• Bioinformatics

November 2020
4
5. MATERIAL AND METHODS
Weka is a data mining application that uses algorithms. Such algorithms can be directly applied
to the data or code-named Java.
"Weka's a platform for:
 Regression.
 Clusters.
 Association.
 Pre-processing data.
 Classification Rating.
 Visualization.”[13]
We have Weka classifiers to approximate measurable or numerical quantities. Decision and lists
are available for learning systems, vector-support machines, case-dependent classifiers, technical
regression, and Bayes networks. If loaded, all tabs are allowed. The best algorithm for basic data
representation can be found for parameters, tests and errors.
The tab Cluster aims to identify clusters or event groups in data set. Clustering provides the
customer details for analysis. Clustering uses the training set, percentage division, test set, and
classes issued, for which users can ignore other requirement-based data set attributes. "K-Means,
EM, Cobweb, X-means and Farthest First in Weka."
6. PROPOSED METHOD:
6.1. Sequential Minimal Optimization (S.M.O.)
The new teaching algorithm supporting vector machines ( SVMs) is minimal sequential
optimization. In 1998, John Platt developed it for minimal sequential optimization (S.M.O.)[14],

November 2020
5
a simple and easy approach for SVM preparation. The underlying theory is to solve the dual
quadratic optimization by maximising the complete two-component sub-set.
S.M.O. 's advantage is the simple and clear introduction. A broad topic of quadratic programming
is understanding a vector-supporting computer.
"S.M.O. distinguishes this core quadratic programming problem in a number of small potentials.
Such a wide square
Analytically solved problem method, which avoids the use of time-consuming quadratic
numerical programming as an internal loop. S.M.O. has a continuous memory feature, allowing
S.M.O. to do intense exercises. Since matrix calculation is avoided, standard S.M.O. scales for
different testing problems like linear and quadratic chunking SVM algorithm scales like linear
and cubic. S.M.O. time is regulated by SVM evaluation; S.M.O. is also the fastest for linear,
small data sets.
7. FRAMEWORK
Fig.3 provides an illustrative outline of SVM and gene-dependent clustering process. Below is a
genetic clustering enhanced SVM process algorithm.
Fig. 3. Proposed Framework of S.M.O. classification with genetic algorithm
Select initial variables indexed by SVM philtre. Output dependent on adaptation, precision, estimation,
memory, accuracy. In the data mining industry, these metric metrics played a crucial role in assessing the
effects of different classifications and directing the algorithms as shown in Table I

November 2020
6
Classification efficiency is calculated in the tables below.
7.1. The Proposed Hybrid Feature Selection Algorithm
Within this section, as our observations indicate, we identify a primary genetic clustering S.M.O.
algorithm. Several operations must be determined before using genetic algorithm. It is the first
population, fitness, selection, and crossover.
STEP 1: Population initialization: randomly generated initial population. A random function (g)
generates a random number in [0 , 1] for each element. If the number exceeds a threshold, g=1.
Otherwise, g=0. For example, the threshold value may be 0.5 to equate 1 or 0.

November 2020
7
STEP 2: Minimal sequential vector recognition interface aid preparation.
STEP 3: It extracts all missing properties globally and converts them into binary attributes. This
functions are also simply standardised. (This is important in the study of output parameters based
on structured data, not original data.
STEP 4: The genetic algorithm integrates genetic algorithm K-means into a modern statistical
clustering method.
8. RESULTS AND DISCUSSION
In our opinion, genetic algorithms are original methods that can be used for the extraction of
functions. They suggest a system for the evaluation of individuals based on the combination of
SVM[15] classifiers that have been qualified for each function. More specifically, we associate
an SVM classifier with each primitive and implement a selection of classifiers. In the method we
propose, classifier learning is Performed one step before each primitive genetic algorithm. The
genetic algorithm fitness function is calculated by a mixture of these classes. We significantly
reduce the training period for the that classifier. Therefore, the proposed selection approach
significantly reduces implementation time. This dataset includes569 reports of breast cancer
cases, 357 of whom were healthy, and 212 of whom were malignant. Class name and 32 function
number are identical to the brain or malignant type of breast cancer. This functions are Measured
by visual representation of the breast mass aspirates needle (F.N.A.) representing the cell nucleus
characteristics in the photo.
8.1. S.M.O. Based SVM Algorithm Results
We run 5 K folded philtre output S.M.O. Classifier. Various other algorithms were contrasted
with our system and can be seen below:
Where, B = Benign, M = Malignant.
It can be shown that S.M.O. algorithm worked better in terms of Accuracy, Recall, F-measure,
and R.O.C. field than other algorithms.

November 2020
8
8.2. Genetic Clustering Algorithm Results
Founded by [15]. Standard Manhattan Distance class is used for outcome similarity calculations.
Easy K-Means incorporates the basic missing qualities. If a mechanism creates a chromosome
with all the records of one cluster, chromosomes are changed until at least two
Here, the algorithm took 38.81 seconds to run.
Ultimately, it was found that the overall wrongly clustered instances is 36, 6.32 percent of the
entire data collection. Visualization of cluster forming can be seen in the figure below indicating
cluster formation selection.

November 2020
9
Blue in the database shows malignant examples, and red in the Map indicates Benign cases. The
plot reveals that clustering was successful.
9. CONCLUSION
This essay presents a new approach to feature collection. This selection approach is based on
genetic algorithms combined with 5k fold-cross-validation SVM classification. Our system was
evaluated on the Wisconsin dataset for breast cancer with several other forms classifier. We
obtained very high accuracy results in grouping and quick results in clustering Breast-cancer
diagnosis. This method not only provides a more accurate model estimation utility estimation, but
also helps to assess the best ML algorithms parameters. Results show the suggested method's
robustness. Our goal is to launch the G.A. immediately. Analysis to ensure the required balance
of consistency and time of diverse datasets from other fields of inquiry.
REFERENCES
1. E.C. E.C. Fear, P.M. Meaney, and M.A.Stuchly,"Breast Cancer Detection Microwaves, "IEEE
Potentials, vol.22, pp.12-18,February-March 2003.

November 2020
10
2. Homer MJ.Mammographic: A realistic solution. Boston, MA, second edition, 1997.
3. Reston VA, Illustrated Breast Imaging Reporting and Data System (BI-RADSTM), third edition,
1998.
4. Urmaliya, Ajay, Singhai. (2013). (2013). Minimal sequential optimization to help vector machine for
breast cancer diagnosis function collection. 2013 IEEE 2nd International Image Processing
Conference, IEEE ICIIP 2013. 481–486. ICIIP.2013. 6707638 10.1109/2013.
5. Bittern, R., Dolgobrodov, D., Marshall, R., Moore, P., Steele, R. Artificial neural cancer networks.
All Hands Conference 19 (2007), pp.251 – 263.
6. Bellaachia, A., AND Guven, E. Estimating breast cancer longevity through data mining.
7. Afef Ben Brahim, Mohamed Limam, 2017, "Ensemble feature selection for high-dimensional data: a
new approach and comparative analysis," IFCS, vol. 12(4), 937-952
8. Huerta, E.B., 2006, "A Hybrid GA / SVM method for gene discovery and microarray data
classification," springer, vol.3907 pp.34–44
9. Sallehuddin, R.N.H.aidillah, S.H., Mustaffa, N.H., 2014, "Liver cancer detection using artificial
neural networks and supporting vector machines" In: Proceedings of the International Conference on
Progress in Communication Network, and Computation, Elsevier Research, pp 487-493.
10. Jabbar, M.A., Deekshatulu, B.L., Chandra, 2012, "Heart disease prediction method using associative
and genetic algorithm."
11. S. S. Nikam, "A Comparative Study of Data Mining Algorithms Classification Strategies," Oriental
Computer Science & Technology Journal, vol.8, pp.13-19, April 2015.
12. S. Neelamegam, E.Ramaraj, "International Journal of P2P Network Developments and Technology
(IJPTT), vol. 4, pp. 369-374, September 2013.
13. Eibe Frank, Mark A. Hall, Witten (2016). Workbench WEKA. "Data Mining: Functional Machine
Learning Methods and Techniques," Morgan Kaufmann, Fourth Edition, 2016.
14. Fourteen. "Platt, J.C.: Minimum sequential optimization: quick algorithm for training vector
machines. MSR-TR-98- 14, Microsoft Research, 1998.
15. Fifteen. 'Islam, M. Z., Estivill-Castro, V., Rahman, M. A. (From 2018). Combining K-Means and a
Genetic Algorithm into a Novel Method for High Quality Clustering. Application-expert systems.
91:402-417 AM.
16. Sangaiah, Vincent Antony Kumar, A. (2019). (2019). Improving the efficiency of medical diagnosis
using hybrid function selection by reliefs and entropy-based genetic search (RF-EGA) approach:
breast cancer prediction. Computing array. 22. 22. 10.1007 / s10586-1702-5.

SVM &GA-CLUSTERING BASED FEATURE SELECTION APPROACH FOR BREAST CANCER DETECTION

More Related Content

What's hot (17)

Similar to SVM &GA-CLUSTERING BASED FEATURE SELECTION APPROACH FOR BREAST CANCER DETECTION (20)

Recently uploaded (20)

SVM &GA-CLUSTERING BASED FEATURE SELECTION APPROACH FOR BREAST CANCER DETECTION