Reduction Attributes on K-Nearest Neighbor

Algorithm (KNN) using Genetic Algorithm

M. Arinal Ihsan

, Muhammad Zarlis, Pahala Sirait

Departement of Computer Science and Information Technology, University of Sumatera Utara, Medan, Indonesia

Keywords: K-Nearest Neighbor, Classification, Genetics, Selection, Rank, KNN.

Abstract: Diabetes mellitus (DM) is a serious health problem both in Indonesia and in the world. Data mining techniques

have been done to help diagnose diabetes. Attribute selection is a process to identify and remove attributes

with irrelevant or excessive values. In this study, attribute selection was performed using genetic algorithm

implemented at K-Nearest Neighbor (KNN) for classification task. The genetic algorithm aims at sorting

attributes by rank where the greater of an attribute the more significant the attribute for the classification task.

The test was performed on Indians dataset of 768 data. From the test, we got the best combination with 1

attribute selection: 3 and 4 attribute reduction from K-Nearest Neighbor (KNN) accuracy before it was

reduced 76,52%, and after reducing 76.96%. While the selection of 2 attributes is : reduce the attributes 1 and

4. The comparison of the results of K-Nearest Neighbor (KNN) accuracy before it is reduced is 76,52%, and

after attribute reduction is 79,57%. These results prove that the comparison of the results obtained attribute

deduction while maintaining the optimization of results before and after eliminate attributes.

1 INTRODUCTION

Diabetes mellitus is a metabolic disease with

characteristics of hyperglycemia that occurs due to

abnormalities of insulin secretion, insulin work or

both. If you have been exposed to chronic diabetes,

then there will be long-term damage, dysfunction or

failure of some organs of the body especially the

kidneys, eyes, nerves, heart, and blood vessels.

Diabetes mellitus (DM) is a serious health problem

both in Indonesia and in the world. According to a

2005 WHO survey, Indonesia as a lower-income

country ranks fourth with the largest number of

people with diabetes mellitus in the world after India,

China and the United States. Based on Indonesia

Health Profile 2008, diabetes mellitus is the cause of

death rank six for all age in Indonesia with the

proportion of death 5,7%, under stroke, TB,

hypertension, injury, and perinatal. Attribute

selection is a process to identify and remove attributes

with irrelevant or excessive values. In this research,

the selection of attributes using genetic algorithm is

implemented in K-Nearest Neighbor (KNN) for

classification task. Genetic algorithm aims to perform

sorting attributes based on rank (rank) where the

greater of an attribute the more significant the

attribute for prediction task.

This selection process is implemented in the K-

Nearest Neighbor (KNN) Method. KNN Is one of the

most commonly used algorithms in the classification

or prediction of new data. The purpose of the KNN

algorithm is to classify new objects based on

attributes and training samples. and KNN algorithm

is a powerful algorithm for training data that has a lot

of noise (Lestari, 2014). And on other studies

(Nirmala Devi, M. Appavu, S., Swathi, U.V., An

amalgam KNN to predict diabetes mellitus :, IEEE,

2013) This study observed that the KNN algorithm

provides significant performance on various datasets.

Which method used for attribute selection in this

research is Genetic Algorithm with selection of

Roulette wheel selection. Roulette wheel selection

Individuals are mapped in a line segment in order so

that each individual segment has the same size as its

fitness size. A random number is generated and the

individual that has a segment within the random

number region will be selected. This process is

repeated until a certain number of individuals are

expected. The authors conducted this observational

study from the international journal "Diagnosis of

Diabetes Mellitus using K Nearest Neighbor

Algorithm" (Krati Saxena1, Zubair Khan2, Shefali

Ihsan, M., Zarlis, M. and Sirait, P.

Reduction Attributes on K-Nearest Neighbor Algorithm (KNN) using Genetic Algorithm.

DOI: 10.5220/0010043203710378

In Proceedings of the 3rd International Conference of Computer, Environment, Agriculture, Social Science, Health Science, Engineering and Technology (ICEST 2018), pages 371-378

ISBN: 978-989-758-496-1

 2021 by SCITEPRESS – Science and Technology Publications, Lda. All r ights reserved

371

Singh3, July 2014). Before doing data processing this

research will perform the process of normalization of

data by using Min-Max method with the aim of

avoiding the occurrence of various anomalies of data

and not consistency of data, the data in this study has

2 data types namely integer and real data so that the

researcher normalize the data. "Min Max

Normalization Based Data Perturbation Method for

Privacy Protection" (Yogendra kumar Jain, santosh

Kumar Bhandre, 2011).

2 GENETIC ALGORITHM (GA)

The genetic algorithm is a computational algorithm

that inspired Darwin's theory of evolution which

states that the survival of a creature is influenced by

the rule that high fitness-worthy individuals will

survive while low-fitness individuals will die

(Goldberg D. E. 1989). Genetic algorithms are

especially applied as computer simulations where a

population of abstract representation (called

chromosomes) of candidate solutions (called

individual) on an optimization problem will develop

into a better solution. Traditionally, solutions are

represented in binaries as string '0' and '1', although it

is also possible to use different encodings

(encodings).

Genetic algorithms use the mechanisms of

natural selection and genetic science so that the terms

in the Genetic Algorithm will be in line with the terms

of natural selection and genetic science in general. A

solution developed in a genetic algorithm is called a

chromosome, whereas the chromosome is called a

population. A chromosome is formed from the

constituent components called a gene and its value

can be a numeric, binary, symbol or character. It is

seen from the problems to be solved. These

chromosomes will evolve continuously which will

later be called a generation. In each generation the

chromosome evaluated the success rate the value of

the solution to the problem to be solved using a

measure called fitness.

2.1 General Structure of Genetic

Algorithm

In general the structure of a genetic algorithm can

define with the following steps: (John Holland, 1975).

a) Determining the initial population, the genes

that fill each chromosome are generated

randomly with the N chromosome where the

population size depends on the problem and

corresponds to the solution domain.

b) The value of fitness, is the value that states

whether or not a solution (individual). This

fitness value is used as a reference in

achieving the optimal value in Genetic

Algorithm. Genetic Algorithm aims to find

individuals with the highest fitness value. In

general, the fitness function is derived from an

objective function with a non-negative value.

If the objective function has a negative value,

it is necessary to add a constant of C so that

the fitness value is not negative.

c) Selection, Selection is used to select which

individuals will be selected for cross breeding

and mutation. Selection is used to get a good

parent candidate. The higher the fitness value

of an individual the more likely to be elected.

d) Crossover Operator This function aims to

cross two chromosomes, resulting in new

chromosomes that carry different characters

(genes). This process is done many times in

the population. The chromosome to be omitted

is determined randomly. The procedure of

crossing one point as follows 1). Determine

the number of populations that will experience

a cross, based on pc.

2). Select two chromosomes as the parent, ie

p1 and p2.

3). Determine the crossover position by

generating random numbers with the range 1

to (length of chromosome-1).

e) Mutations, which play a role in replacing the

missing genes of the population due to a

selection process that allows the re-emergence

of genes that do not appear in the initial

population. Mutations can be made of all the

genes present with a given mutation

probability. If the random number generated is

less than the probability of the specified

mutation then change the gene to its inverse

value. In simple Genetic Algorithms, mutation

probability values are fixed during evolution.

The mutation (Pm) opportunity is defined as

the percentage of the total number of new

genes that will be raised for evaluation. If the

mutation chance is too small, many possible

useful genes are never evaluated. But if the

chance of this mutation is too great, it will be

too much random noise, so it will lose the

resemblance of the parent. Other mutation

processes can be mutated in genes as much as

the probability of a mutation * the number of

genes, in which the gene position to be carried

out by mutation is randomly selected (John

ICEST 2018 - 3rd International Conference of Computer, Environment, Agriculture, Social Science, Health Science, Engineering and

Technology

372

Holland of the University of Michigan in

1975).

3 K-NEAREST NEIGHBOR (KNN)

The K-Nearest Neighbor (K-NN or KNN) algorithm

is a method for classifying objects based on learning

data closest to the object. Learning data is projected

into many-dimensional space, where each dimension

represents a feature of the data. This space is divided

into sections based on the classification of learning

data. A point in this space is marked as class C if class

C is the most common classification of the nearest

neighbor of the titk. Close or away neighbors are

usually calculated based on Euclidean distance

(Kozma, Lazszlo 2013).

In the learning phase, this algorithm only stores

the features and classification vectors of learning

data. In the classification phase, the same features are

calculated for the test data (the classification is

unknown). The distance from this new vector to all

vectors of learning data is calculated, and a number

of k kits closest are drawn. The new point of

classification is predicted to include the most

classification of the points.

The best value of K for this algorithm depends

on the data; generally, a high k value reduces the

noise effect on the classification, but makes the

boundary between each classification more blurred.

A good K value can be selected with parameter

optimization, for example by using cross-validation.

A special case in which classification is predicted

based on the closest learning data (in other words, K

= 1) is called the nearest neighbor algorithm

(Mohammed J. Islam, 2011).

Steps to calculate KNN algorithm method:

a. Specifies the parameter K (the closest

number of neighbors).

b. Calculates the quadratic distance of Euclid

(query instance) of each object against the

given sample data. With

𝐷



𝑥,𝑦







𝑥



𝑦











(1)

Keterangan:

𝑥𝑖 = Sampel Data

𝑦𝑖 = Data Uji / Testing

𝑖 = Variabel Data

d = Jarak

c. Then sort those objects into groups that have

small Euclid spacing. Incorporate category

Y (Klasifikasi Nearest Neighbor)

The accuracy of the K-NN algorithm is strongly

influenced by the presence or absence of irrelevant

features, or if the weight of the feature is not

equivalent to its relevance to the classification.

Research on this algorithm largely discusses how to

choose and weight the features, so that classification

performance gets better.

Previously there were several related studies on

the use of the KNearest Neighbor method. As

research conducted by Khatib Alkhalit et al with the

title "Stock Prediction Using K-Nearest Neighbor

(KNN) Algorithm". In the study, stock prices are

predicted because the stock market is considered as a

field of trade that provides an easy profit

with a low level of risk and benefits for investors,

management and decision makers in determining

investment decisions.

The K-Nearest Neighbor method is chosen

because it has high accuracy and ratio

small mistake. The dataset used in the research is

data of the stock exchanges of five major companies

listed in the Jordanian stock exchange period of juni

2009 until december 2009. From result of research

done got predicted results with high accuracy and

results close to price actual stock.

There is also research on the use of K-Nearest

method Neighbor is implemented on prediction.

Research by title "Realtime Highway Trafic Accident

Prediction Based on k-Nearest Neighbor Method ".

Problems discussed in the research that is often the

occurrence traffic accidents and obstacles in

identifying traffic accidents in realtime. The purpose

of the research is to predict the potential traffic

accidents by identifying normal traffic conditions and

dangerous traffic conditions with realtime data.

4 DATA REDUCTION

Data reduction is a process of selection, focusing on

simplification, abstraction, rough data transformation

arising from field notes (Miles and Huberman (1992:

16)).

The steps taken are to sharpen the analysis,

classify or categorize into each problem through brief

descriptions, directs, discards unnecessary, and

organizes the data so that it can be withdrawn and

verified. Reduced data include all data on research

problems.

Reduced data will provide a more specific picture

and make it easier for researchers to collect further

data and seek additional data if needed. The longer

the researchers are in the field the more data the more,

the more complex and complicated. Therefore, data

reduction needs to be done so that the data does not

overlap so as not to complicate further analysis.

Reduction Attributes on K-Nearest Neighbor Algorithm (KNN) using Genetic Algorithm

373

4.1 Presentation of Data

After the data is in the reduction, the next step of

analysis is the presentation of the data. Presentation

of data is as a set of arranged information that

provides the possibility of drawing conclusions and

taking action. (Miles and Huberman, 1992).

The presentation of data is directed to reduce the

result data is organized, arranged in a relationship

pattern so that more easily understood. Presentation

of data can be done in the form of narrative

descriptions, charts, relationships between categories

and flowcharts. Presentation of data in such forms

facilitate researchers in understanding what happens.

In this step, the researcher tries to arrange the relevant

data so that the information obtained is concluded and

has a certain meaning to answer the research problem.

The presentation of good data is an important step

towards achieving a valid and reliable qualitative

analysis. In presenting the data not merely describe in

narrative, but accompanied by a process of

continuous analysis until the process of drawing

conclusions. The next step in the process of

qualitative data analysis is to draw conclusions based

on the findings and verify the data.

4.2 Normalization

Normalization (Normalization) is Process to organize

files to remove repetitive element groups ". The

concept and technique of normalization was

introduced by Dr.E.F Codd in his paper in 1970 and

1972. In his paper, E.F. Codd defines a new data

structure called the relationship data structure

(relational data structure). The term relationship data

indicates a data structure that has relationships with

other data elements, either in one or in other files. In

this normalization process there are several methods

that can be used such as: (J. Han, and M. Kamber,

2006,).

5 RESEARCH METHODOLOGY

The methodology of this research is shown in Figure

Figure 1. Flowchart The KNN Process With and Without

Attribute Selection

Figure 1 is an illustration of the flow of the

methodology of this study describing the steps to be

performed in the study, ranging from the dataset used

in performing attribute selection analysis on KNN

algorithm in predicting diabetes and examples of

research calculations.

5.1 Used Dataset

In this study, the secondary data used were taken from

the UCI Indian Pima database repository. The Pima

Indian dataset consists of 768 clinical data.

All patients are female and are at least 21 years of age

living in phoenix, Arizona, USA. This data set

contains two classes that are presented in binary

variables that are either '0' or '1'. The number "1" of

the test results indicates positive for diabetes and "0"

indicates negative diabetes. The dataset contains 768

patients with 9 numerical variables. There were 268

(34.9%) positive cases of diabetes with marked

classes "1" and 500 (65.1%) cases in grade "0". The

dataset does not contain missing values. Five patients

had glucose "0", 11 patients had BMI "0", 28 patients

ICEST 2018 - 3rd International Conference of Computer, Environment, Agriculture, Social Science, Health Science, Engineering and

Technology

374

had blood pressure "0", 192 patients had skin fold

thickness "0", 140 had serum insulin level "0". The

attributes of the pima indian dataset are shown in

table 1.

Table 1: Attribute Dataset Diabetes Mellitus in Indian

Pima

attrib

ute

numb

Attribute

Name

Descriptio

Type unit

PREGNAN

Number of

time

pregnan

Numeric

A2 GTT

2-hours

OGTT

plasma

glucose

Numeric

mg/dl

A3 BP

Diastilic

blood

preseure

Numeric

mmH

SKIN

Triceps

skin fold

thickness

Numeric

A5 INSULIN

2-hours

serum

insulin

Numeric

U/ml

A6 BMI

Body mass

index

(kg/m)

Numeric

Kg/m

A7 DPF

Diabetes

pedigree

function

Numeric

AGE

Age of

patient

(years)

Numeric

Class DIABETES

Diabetes

onset

within

years (0,1)

Numeric

5.2 Pre-processing and Sampling

Stages

This stage is done to get the data clean and ready to

use. The pre-processing data stage includes

identification and attribute identification and

selection, handling missing values, and value

discretization process. Statistical analysis of Indian

Pima diabetes dataset is shown in table 2 and table 3

below. The average value before normalization in

table 2. Furthermore the data is normalized using the

equation:







∗











𝑛𝑒𝑤_𝑚𝑖𝑛 (2)

The equation above to normalize the data.

Normalized results in table 3.

Table 2: Before Normalization

Atribut

Mean Standard

Deviation

Atr

1 3.84 3.37

Atr

2 120.89

31.97

Atr

3 69.1

19.35

Atr

4 20.53 16.0

Atr

5 79.79 115.24

Atr

6 31.99 7.88

Atr

7 0.47 0.33

Atr

8 33.24

11.76

Table 3: After Normalization

Atribut

Mean Standard

Deviation

Atr

1 0.226 0.19

Atr

2 0.608 0.16

Atr

3 0.566 0.15

Atr

4 0.207 0.16

Atr

5 0.094 0.13

Atr

6 0.477 0.11

Atr

7 0.168 0.14

Atr

8 0.204 0.19

6 RESULTS AND DISCUSSION

For this dataset, the results of testing with data

training as much as 538 data and data testing as much

as 230 data. By conducting trials on all attributes then

obtained the correct data as much as 445 of 538 data.

The equation of KNN classification accuracy using

the following equation:

accuracy

    

   

∗ 100% 3

Based on equation 2 obtained the accuracy of test

results shown in the table below. By using the above

accuracy equation for k = 10 then obtained the result

of classification calculation accuracy using KNN of

82,71%.

Table 4: Result of KNN Accuracy Testing

The value of

K is used

Data True

Accuracy

Results

k-5 430

79,93%

-10 445

82,71%

k-15 408

75,84%

k-20 428

79,55%

Reduction Attributes on K-Nearest Neighbor Algorithm (KNN) using Genetic Algorithm

375

k-25 405

75,28%

k-30 413

76,77%

k-35 404

75,09%

We can conclude from table 4 above k-10 is the

highest accuracy obtained by the accuracy of 82,71%.

So for the next process for taking the accuracy value

will use k = 10.

6.1 Selection Calculation Results 1

Attribute On KNN Classification

Using GA

Then we calculate the value of KNN accuracy that has

been in the reduction by using data testing for testing

as much as 230 Dataset. Of the 8 attributes we have

will do a combination of 8 using the Factorial formula

𝐶







!

! !

= 8 combination.

In combination of 8 above attributes by using

K = 10 Obtained the highest fitness result is 76,96%.

As in table 5 below:

Table 5: The result of selection of 1 attribute

No Attribute

Combination

True percentage

C1 1,2,3,4,5,6 &

171 74,34%

C2 1,2,3,4,5,6 &

171 74,34%

C3 3 & 8 139 60,43%

C4 1,2,3,4,5,7 &

176 76,39%

C5 1,2,3 & 7 170 73,91%

C6 1,2,5,6,7& 8 177 76,96%

C7 2,3,4,5,6 & 8 176 76,39%

C8 1,3,4,6,7 & 8 164 71,30%

So it can be concluded from table 5 above the

highest accuracy value on the 6th chromosome that is

equal to 76.96% by reducing attributes 3 and 4.

Next look at the comparison before and after

reducing attributes in table 6 and table 7 below:

Table 6: Accuracy Results Before Reduction

Description Accuracy True Data

Amoun

KNN before

reduction

Attributes

76,52% 176

Table 7: Accuracy Result Before Reduction

Description Accaruracy True Data

Amoun

KNN after

reduction

Attributes

76,96% 177

Furthermore, for comparison of the correct level

of data similarity between the classification before

and after reducible can be seen in table 8 below:

Table 8: The result of comparison of data similarity

classification

Classification Results

Accuracy

of Data

Similar

KNN before

attribute

reduction

KNN after the

combined

reduction of 1

attribute with

84,78%

the table above shows the result of the similarity

between the classification KNN before and after

reducible obtained 84.78%, obtained the correct data

as much as 195 of 230 data.

6.2 Selection Calculation Results 2

Attribute On KNN Classification

Using GA

Of the 8 attributes we have will do a combination of

8 using the Factorial formula .

𝐶







!

! !

= 28 Kombinasi

In the combination of 28 attributes above by using

K = 10 Obtained fitness results highest ie 79.57%. As

in the table below:

Table 9: The result of the selection of 2 attributes

No Attribute

Combination

True percentage

C1 3,4,5,6,7 & 8 161 70%

C2 3,4,5,6, 7 & 8 161 70%

C3 1,4,,6 & 8 149 64.78%

1,2,3,4,5,6 &

168 73.04%

C5 1,2,4,5 & 7 177 76.96%

C6 4,5,6,7 & 8 166 72.17%

ICEST 2018 - 3rd International Conference of Computer, Environment, Agriculture, Social Science, Health Science, Engineering and

Technology

376

C7 2,4,5 & 6, 161 70%

C8 2,3,5 & 7 166 72.17%

C9 1,3 & 7 147 63.91%

C10 2,3,4,6 & 8 176 76.52%

C11 1,3 & 4 152 66.09%

C12 3,4,5 & 7 162 70.43%

C13 1,2,5,7 & 8 182 79.13%

C14 1,2,3,5,7 & 8 172 74.78%

C15 2,3,5,6 & 8 180 78.26%

C16 3,4,6,7 & 8 162 70.43%

C17 3,5,6,7 & 8 167 72.61%

C18 1,2,4,5 & 8 176 76,52%

C19 2,3,5,6,7 & 8 183 79,57%

C20 2,3,4,5 & 7 171 74.35%

C21 1,2 & 4 165 71.74

C22 1,4,6 & 8 149 64.78%

C23 1,4,6,7 & 8 162 70.43%

C24 1,2,3,4 & 6 171 74.35%

C25 3,5,6,7 & 8 167 72.61%

C26 4,5 & 8 156 67.83%

C27 1,3,6,7 & 8 164 71.30%

C28 3,4,6,7 & 8 162 70.43%

So it can be concluded from table 9 above the

highest accuracy value on the 19th Chromosome that

is equal to 79.57% by reducing attributes1 and 4.

Next look at the comparison before and after

reducing attributes in table 10 and table 11 below:

Table 10: Accuracy Results Before Reduction

Information Accuracy

True Data

Amoun

KNN before

reduction

Attributes

76,52% 176

Table 11: Accuracy Result After Reduction

Information Accuracy

True Data

Amoun

KNN After

reduction

Attributes

79,57% 183

Furthermore, for comparison of the correct level

of data similarity between the classification before

and after reducible can be seen in table 8 below:

Table 12: hasil perbandingan klasifikasi kemiripan data

Classification Results

Accuracy

of Data

Similar

KNN before

attribute

reduction

KNN after the

combined

reduction of 1

attribute with

87,39%

the table above shows the results of similarities

between the classification KNN before and after

reduced 87.39% obtained, obtained true data as much

as 201 of 230 data

7 CONCLUSION

Based on the testing and evaluation of KNN method

in attribute selection using genetic algorithm, the

conclusion obtained by the selection of 1 attribute

accuracy are: 76,52% before attribute reduction and

76,96% after reduction by reducing attribute 3 and 4,

and the result of classification the data likeness rate is

84.78%. While the result of accurate selection of 2

attributes are: 76,52% before reduction and 79,57%

after reduction by reducing attribute 1 and 4, and

result of classification of data likeness level equal to

87,39%.

The result of the accuracy percentage of the

combination of 1 attribute selection increased by

0.44% compared to before the reduction. And the

accuracy of the 2 attribute selection combination rose

by 3.05% before reduction. This proves that the

process of attribute selection has increased the

accuracy in determining the classification. And the

more combinations it performs the better the

optimization results of the genetic algorithm.

REFERENCES

Arwa Al-Rofiyee, Maram Al-Nowiser, Nasebih Al Mufad,

“Integrating Decision Tree and ANN for

Categorization of Diabetics Data” International

Conference on Computer Aided Engineering,

December 13-15, IIT Madras, Chennai, India (Asha

Gowda Karegowda , M.A. Jayaram, A.S. Manjunath

(2012)

Gorunescu, F, “Data Mining Concepts, Models and

Techniques”, Verlah Berlon Heidelberg:Spinger,

2011.)

T. Karthikeyan, “An Analytical Study on Early Diagnosis

and Classification of Diabetes Mellitus” (International

Reduction Attributes on K-Nearest Neighbor Algorithm (KNN) using Genetic Algorithm

377

Journal of Computer Application (2250-1797) Volume

5– No. 5, August 2015, )

Krati Saxena1, Zubair Khan. Diagnosis of Diabetes

Mellitus using K Nearest Neighbor Algorithm”, Shefali

Singh3, July 2014)

Lestari, Mei, 2014. “Penerapan Algoritma Klasifikasi

Nearest Neighbor (K-NN) untuk mendeteksi penyakit

jantung”, Faktor Exacta 7(4): 366-371, ISSN:1979-

276X.

Depkes RI, “Profil Keseheatan Indonesia”, Jakarta:

Depkes RI, 2009.

J. Han, and M. Kamber, “Data Mining: Concepts and

Techniques”, Second edition, 2006, Morgan

Kaufmann, USA.

Miles, B. Mathew dan Michael Huberman. 1992. Analisis

Data Kualitatif Buku Sumber Tentang Metode-metode

Baru. Jakarta: UIP

Min Max. Normalization Based Data Perturbation Method

for Privacy Protection” (Yogendra kumar Jain, santosh

Kumar Bhandre, 2011).

http://www.psu.edu.sa/megdam/sdma/Downloads/Post

ers/Pos ter%2003.pdf, diakses tanggal 16 April 2016)

Penelitian (Nirmala Devi, M.; Appavu, S., Swathi, U.V.,

―An amalgam KNN to predict diabetes mellitus:,

IEEE, 2013)

Setiawan Meddy, “Buku Ajar Endokrin, Malang, FK:UMM

Sitompul, O.S., 2008, Data Warehouse dan Data Mining

untuk Sistem Pendukung manajemen, Pidato

Pengukuhan Jabatan Guru Besar Tetap dalam Bidang

kecerdasan Buatan, Fakultas MIPA USU, Medan.

Bennett K, Henman M trinity College Dublin, Dublin,

Ireland 2015. The Impact of Self-Monitoring Of Blood

glucose (SMBg) On Prescription Costs In newly

Treated Type 2 Diabetes Melitus (T2DM): a

retrospective Cohort Analysis Grimes RT.

Z. Zukhri, Algoritma Genetika Metode Komputasi

Evolusioner untuk Menyelesaikan Masalah Optimasi,

Yogyakarta: ANDI, 2014.

ICEST 2018 - 3rd International Conference of Computer, Environment, Agriculture, Social Science, Health Science, Engineering and

Technology

378