JURNAL TECH-E - VOL.
1 NO.
Available Online Version: http://jurnal.
id/index.
php/te JURNAL TECH-E | 2581-1916 (Onlin.
| 2598-7585 (Ceta.
| Artikel Data Mining Implementation on Choosing Potential Customers Using K-Means Algorithm on PT.
Koba Metal Indonesia Sandi Kristianto1.
Yusuf Kurnia2 Universitas Buddhi Dharma.
Teknik Informatika.
Banten.
Indonesia Universitas Buddhi Dharma.
Teknik Informatika.
Banten.
Indonesia
SUBMISSION TRACK
Received : February 1, 2018
Final Revision: February 6, 2018
Available Online: February 27, 2018
KEYWORD
Potential Customers.
Data Mining.
Clustering.
K-Means.
Rapidminer.
CORRESPONDENCE
Telepon: 081281500393 E-mail: sandi.
kristianto@gmail.
A B S T R A C T
PT.
Koba metal Indonesia.
is one of roll-reforming cooperations who produce light-steel stuffs which is growing rapidly nowadays.
One of the important thing in customer management is how a cooperation be able to preserve their the effort of preserving customers becomes important for PT.
Koba metal Indonesia.
considering of plenty companies who commits at the same sector.
To prevent the displacement of customers, knowing the potential group of customers is important, so that the company could preserve those potential customers by giving excellent service, etc.
implication of data mining could assist the company to analize the received data from sales transaction to gain potential customers data.
Therefore, a designed application which could implement the data mining for choosing potential customers by clustering and algorithm K-means method is arranged.
Then, the information performes with groups who is categorized into potential customers.
Besides, rapminder application is also used to examine the dataAos accuracy of this built application design.
Hereinafter, this application design is expected to assist companies to choose their potential customers and preserve them to advance their business.
INTRODUCTION
One of the important things of customer management is how a company be able to preserve their owned customers.
In this case.
PT.
Koba metal Indonesia.
as a roll-forming cooperate or the proses of building light-steel into ready-to-use utensils in development section, starts to realize to preserve their customers, considering on the growth of rollforming business who sells same products as To prevent the displacement of customers, knowing the potential group of customers is important, so that the company could preserve those potential customers by giving excellent service or prizes.
The prizes could be souvenirs that given to their customers annually.
PT.
Koba metal SANDI KRISTIANTO / JURNAL TECH-E - VOL.
NO.
Indonesia.
is one of the companies that realize the importance of the connection between loyal customers and the success of companyAos METHODS K-means is an algorithm that be used in partial classification that separate datas into different groups.
This algorithm is able to minimalize the gap between data to its cluster.
Basically, the application of this algorithm in the clustering process depends on the received datas and the conclusions that expected to be achieved in the end of the process.
So in kmeans algorithm application, there are precepts, such as:
How many clusters that needed to be Only have numeric type attribute.
Basically k-means algorithm only takes a piece of those plenty components that is received to become the center of the beginning cluster.
After that, the k-means algorithm will examine each of the component in that data population and mark that component into one of defined cluster center depend on the minimum gap between clusterAos center to another center.
After that, the clusterAos centerAos position will be counted until all of the data components is classified into each of clusters and at last, new cluster will be built.
Data mining is a data analyse to discover an obvious relation and conclude the unknown with the current method that useful and understood by the owner.
Clustering or classifying is a method that is used to divide data circuits into some groups based on their similarity which has determined before.
Cluster is a group of similar data in the same, cluster, and dissimilar to other clusterAos object.
Object will be classified into one or more clusters, so that objects in a cluster have substantial similarity between one and another.
II.
RESULTS
The image below is an applied k-means algorithm methodAos layout on rapidminer.
This process starts in Read excel step, which is an excel files in clustering data are processed with k-means algorithm, after that, these datas will be on apply model stage, to adjust these k-means algorithm clusterised data to enter performance stage.
performance stage, these datas will be processed to produce PerformaceVector, cluster model, and example set output.
Image 1: Algorithm K-Means Application on Rapidminer SANDI KRISTIANTO / JURNAL TECH-E - VOL.
NO.
Image 2: Cluster Model on Rapidminer Gambar 3: K-Means Diagram on Rapidminer Gambar 4: Cluster Model on Rapidminer SANDI KRISTIANTO / JURNAL TECH-E - VOL.
NO.
Table 1.
Comparison Value on 3 Algorithm Methods These variables consist of 3 fuzzy compilations, such as low, medium, and high, which is showed on image 2 and 3.
Each of compilationAos affliationAos functions are formulated as follows: Based on table above, can be known that Nayve Bayes accuracy value is 90.
06%, 84,30% for C4.
5, and 93% for K-Means.
These 3 methods above examine PT.
Kobe metal Indonesia.
Aos sales For K-Means accuracy value is obtained by manual calculation with the formula as From algorithm k-means processAo resultAos data, could be indicated 9 on potential customer category, 1 for less potential customers, and 436 for not potential Compared with the raw data .
ave not been processed with k-means algorith.
has been predicted 16 potential customers, 20 less potential customers, and 410 not potential These type of data later gained the values that will be used to count the accuracy value, as the table follows:
Table 2.
Accuracy Point Table 3.
Sales Data Sample on PT.
KMI
DISCUSSION
Decision making is started by cluster determination stage, which is done based on the aim of data mining process.
On this case, data that is aimed to be generated is potential customerAos data, with 3 categories, such as potential, less potential, and not potential.
Can be concluded that the built clusters are 3/c=3.
the cluster central point could be decided independently or by find out the hight value .
, average value.
, and lowest value .
Here is data example with 40 datas taken.
Cluster center point:
C1=highest value .
hopping frequency, total C1=.
,83664.
SANDI KRISTIANTO / JURNAL TECH-E - VOL.
NO.
C2=average value .
hopping frequency, total C2=.
15,20078.
C3=lowest value .
hopping frequency, total C3=.
,256.
Calculate the gap between data to cluster central point.
Table 4.
Iteration 1 Process Result So the new C3 center is = 2.
2,4318698.
After getting the new cluster result, count the gap between data and the new cluster centerAos point, and categorize again the cluster.
Repeat those stage all over again until clusterAos position on the cluster grouping stage will not change anymore.
Table 5.
Iteration 2 Process Result Calculate the cluster center again with the current clusterAos membership.
The new cluster center is the average of all datas/objects in certain group.
Finding the new C1 So the new C1 center is = 5.
4,64194988
So the new C2 center is = 3.
4,22708541.
Table 6.
Iteration 3 Process Result SANDI KRISTIANTO / JURNAL TECH-E - VOL.
NO.
cluster 1 central point is on .
3,60017.
So that could be cluster central point concluded that cluster 1Aos customers are potential customers, for cluster 2Aos central point is on .
9,26073.
categorized as less potential customers, and for the cluster 3Aos central point is on .
3,6977.
categorized as not potential customers.
Table 7.
Iteration 4 Process Result Because on the 3rd and the 4th iteration has no change of the cluster postion, then the process is stopped.
It is known that on the 4th iteration.
IV.
CONCLUSION
Based on the research that has been done, it can be concluded as follows:
Agglomeration of the kmi co.
Aos sales data could be done by clustering model, with agglomerating those datas into categories basen on dataAos similarity in a category.
To get potential customers data by kmeans algorithm is done by these stages as follows:
Decide the clusterAos amount.
Decide the clusterAos center point.
Calculate the gap between data to clusterAos center point.
Agglomerate data into clusters depend on the shortest gap or the Repeat step b-d, and compare the data position on every result.
If the data position changes, repeat step b-d all over again.
If the position co not change anymoe, the process is done.
The determination is seen by the biggest clusterAos center poin from the built on the last stage when the data position on the cluster stays Getting potential customer data with Rapidminer application is done by importing the first data to be processed can be in the form of an excel or csv module read excel if excel file and read csv if shaped csv file.
The data processed includes the name of the customer .
, shopping frequency, and total after that input K-Means algortima module and also change the parameters such as the number of clusters Then connect the module read SANDI KRISTIANTO / JURNAL TECH-E - VOL.
NO.
excel / csv to K-means algorithm module and forwarded to the endpoint that is already available.
And if it is done, the next data can be processed by Rapidminer application that produces clusters with data that has been grouped and also displays the highest cluster central point value so that it can be determined which cluster into the category of potential SANDI KRISTIANTO / JURNAL TECH-E - VOL.
NO.
REFERENCES