p–ISSN: 2723 - 6609 e-ISSN: 2745-5254
Vol. 5, No. 9 September 2024 http://jist.publikasiindonesia.id/
Jurnal Indonesia Sosial Teknologi, Vol. 5, No. 9, September 2024 3415
Cluster Analysis and Forecasting on Local Shoe Products:
Case Study for Ventela in Indonesia
Vies Sata Zullah
1*
, Raden Mohamad Atok
2
Institut Teknologi Sepuluh Nopember, Indonesia
Email:
1*
2
*Correspondence
ABSTRACT
Keywords: k-means
cluster; exponential
smoothing; ventela shoe
models; e-commerce
transactions; best-selling
cluster forecasting.
Shoes are a secondary need that is in demand by all age
groups. Each shoe brand has many models, which means a
store must provide complete stock to meet consumer needs.
The available models have different purchasing power or
demand, which creates difficulties for stores in determining
shoe products that are often sold and shoe products that are
not in demand by customers. The data used in cluster
formation includes three variables recorded and collected
from e-commerce transactions, namely Number of Visitors,
Number of Buyers, and Total Sales. Based on these
variables, Ventela shoe models are grouped into three
clusters, namely low-selling, normal, and best-selling. Next,
the variable number of transactions for Ventela shoe models
in the best-selling cluster is taken to be predicted using the
Exponential Smoothing method. The forecasts obtained are
used to determine future demand to maximize profits. Based
on the results of the clustering analysis, it was found that the
number of shoe models included in the best-selling cluster
was six, including (1) Ventela Ethnic Low All Black, (2)
Ventela Ethnic Low Black Natural, (3) Ventela Public Low
Black Natural, (4) Ventela Public Low Cream, (5) Ventela
Republic Low Black Natural, and (6) Ventela Republic Low
White. Referring to the sample of this study, which only
spanned less than three years, several shoe models produced
a forecast value of zero. It means, based on the forecast
results for the next 12 months, there may be no sales.
Introduction
Locally made shoes are increasingly appearing for sale on social media and various
e-commerce sites. The number of enthusiasts has also increased significantly. Tokopedia
recorded that sales transactions for local shoe brands have almost doubled. This is driven
by the increasing level of awareness of Indonesian consumers who are proud to use local
products. Apart from that, many people also really appreciate locally-made shoe products
(Yusditara et al., 2022). E-commerce is a platform to make shopping easier for customers.
E-commerce itself facilitates the ability to make transactions from anywhere; customers
Vies Sata Zullah, Raden Mohamad Atok
Jurnal Indonesia Sosial Teknologi, Vol. 5, No. 9, September 2024 3416
can also make purchases directly, there are cuts in distribution channels, and there are
cost savings (Alwiyah & Gata, 2019).
The increasing interest in local shoes has become a concern for local shoe sellers to
pay more attention to the inventory or stock of each model (or what can be called articles).
Some inventory problems that often occur are excess stock (overstock) or running
out/short of stock (stockout) (Rachmawati & Lentari, 2022). Overstock or stockout can
cause several losses. When there is too much stock, it causes reduced profitability due to
the accumulation of goods. This equates to tying up cash that can be used to buy stock
that comes out more frequently. Apart from that, excess stock can also pressure sellers to
sell their products at prices below the margin to get rid of excess stock.
Meanwhile, stock shortages can lead to lost sales opportunities because customers
cannot buy the products they need and result in loss of income for sellers. When
customers cannot buy the products they need, it can cause dissatisfaction and potentially
damage a business's reputation (Rachmawati & Lentari, 2022). Overstock and stockouts
often occur in shoe products that have quite a lot of articles. There are many shoe brands
in Indonesia, one of which is Ventela which is a locally made shoe and is currently quite
famous in Indonesia.
Ventela is a local shoe brand with a casual type, which was introduced in 2017 by
William Ventela, a vulcanized shoe factory owner, in 1989 in Bandung, West Java. With
abundant resources, Ventela Shoes can produce canvas shoes in large quantities and of
the best quality so that all groups can have high-quality shoes at affordable prices. Ventela
does not have an official store that sells shoes retail to end-consumers but instead
distributes them to resellers. So, resellers play an important role in retail sales. There are
quite a lot of articles published by Ventela; there are more than 100 shoe articles. Each
article has different selling power, so we need to know which articles should have more
stock and which are not in demand by the market. These articles need to be grouped or
formed into clusters using k-means clustering and forecasting demand so as not to
disappoint customers and also allow shoe turnover to dash.
Clustering is one of the main methods for organizing a set of data into clusters so
that the elements in each cluster have similarities and differences with other clusters
(Pérez-Ortega et al., 2019). This clustering is used to create a report regarding the general
characteristics of the groups formed, including shoe models ranging from those that sell
poorly, normally and best-selling. One of the most widely used clustering algorithms
currently is K-means because of its ease of interpreting the results and implementation
(Pérez-Ortega et al., 2019).
Previous research related to clustering using the K-Means method was carried out
by (Pratiwi & Marizal, 2022) with the title Application of "K-means Algorithm for
Grouping and Least Square Method for Predicting Goods Sales (Case Study: Buana Mart
Kendari)". This research obtained results that the grouping of initial and sold stock, as
well as predictions of goods sold at the Buana Mart Kendari Store, were successfully
developed by applying the K-Means algorithm and the Least Square method. Apart from
Cluster Analysis and Forecasting on Local Shoe Products: Case Study for Ventela in Indonesia
Jurnal Indonesia Sosial Teknologi, Vol. 5, No. 9, September 2024 3417
that, (Jananto, 2022) have also conducted research using the ¬K-Means method with the
title "Application of Data Mining to Product Sales Using the K-Means Clustering Method
(Case Study of Kakikaki Shoe Store)". This research explains that the advantages of using
the K-Means Clustering method are that it is easy to run, fast, and flexible, and the results
are easy to understand and can be explained to many people.
In this research, the demand for Ventela shoe models in the best-selling cluster was
predicted using the Exponential Smoothing method. The exponential smoothing method
is a forecasting method for time series data by giving weight to previous data to predict
the value of the following data. This model was chosen because it can accommodate
stationary, trend and seasonal patterns simultaneously, even if the amount of data
(sample) is limited. There are three types of exponential smoothing methods, namely
single, double and triple. Single is used for data that has a stable fluctuating pattern,
double is used for data that has a trend pattern, and triple is used for data that has a trend
and seasonal pattern (Maricar, 2019).
Fahrudin et al. (2022) carried out previous research that applied the exponential
smoothing method, titled "Demand Forecasting of Automobile Sales Using Least Square,
Exponential Smoothing, and Double Exponential Smoothing." (Zunaidi, 2022) also
conducted research related to "Analysis of Inventory Management in Order to Reduce
Overstock (Case Study of TVF Footwear)." This research uses the Exponential
Smoothing and Trend Analysis method to predict demand for the three research products.
Based on the previous description, researchers want to conduct research like
(Zullah, 2024) using k-means to obtain clusters of Ventela shoe models. However, the
Ventela shoe models from one of the cluster results (bestselling) obtained will be
predicted using both methods using exponential smoothing to produce demand forecasts
for the next period.
Research Methods
K-Means Cluster
K-Means is the simplest and most common clustering method. It can group quite
large amounts of data with fast and efficient computing time (Santosa, 2007). The
clustering process begins by identifying data with the Euclidean formula, as in Equation
(1), and is illustrated in Figure 1.
Figure 1 Euclidean formula illustration
Based on Figure 1, in general the Euclidean formula can be stated as follows.
(1)
Vies Sata Zullah, Raden Mohamad Atok
Jurnal Indonesia Sosial Teknologi, Vol. 5, No. 9, September 2024 3418
Where is the distance from point to point ,
indicates the -th point ,
indicates the center point of the -th cluster , and indicates the number of attributes.
A data becomes a member of the kth cluster if the distance of the data to the centre of the
kth cluster is the smallest compared to the distance to the centres of other clusters. Next,
form groups of data that are members of each cluster. The new cluster centre value can
be calculated by finding the average value of the data that is a member of the cluster using
the formula in Equation (2).
(2)
Where
is the centroid point of the -th cluster,
shows the amount of data in
the -th cluster, and
is the -th data in the -th cluster
In general, the K-Means Cluster initializes centroids randomly. According to
(Dwiarni & Setiyono, 2020), the steps for clustering using the K-Means Cluster method
are as follows:
1. Select the number of clusters .
2. Initialize k cluster centers. Most often, this is done randomly.
3. Allocate all data/objects to the nearest cluster. The distance between them determines
the closeness of two objects, and the distance between the data and the cluster centre
determines the proximity of data to a particular cluster.
4. Recalculate the cluster centre with the current cluster membership. Reassign each
object using the new cluster centre. If the cluster centre does not change again, then
the clustering process is complete.
One method commonly used to determine the number of clusters is the elbow
method. By the elbow method, it can be observed how the sum squared error (SSE) value
changes as the number of clusters increases. In the SSE graph against the number of
clusters, one can look for points where the decrease in inertia slows down and resembles
an elbow shape, which can provide clues about the optimal number of clusters. This point
indicates that adding clusters afterwards does not provide significant benefits in reducing
SSE (Shi et al., 2021).
Data
The data used in this research is secondary data obtained from historical sales
transactions carried out by one of the Ventela shoe resellers in Surabaya. Sales data used
starts from August 2021 to March 2024. This research applies two statistical methods,
namely clustering and forecasting. Therefore, there are two different variable definitions
Cluster Analysis and Forecasting on Local Shoe Products: Case Study for Ventela in Indonesia
Jurnal Indonesia Sosial Teknologi, Vol. 5, No. 9, September 2024 3419
according to each method. The following research variables used in this research are
explained in Table 1 and Table 2.
Table 1
Variables for the clustering
Variable
Symbol
Unit
Information
Number of
visitors
People
The number of individuals who access
product pages on e-commerce.
Number of
buyers
People
The number of individuals who
purchased the product.
Total sales
Rupiah
Total sales are income generated from
e-commerce transactions.
Table 2
Variables for the forecasting
Variable
Type
Symbol
Unit
Information
Number of
product sold
Dependent
Piece
Number of products sold on e-
commerce.
Analysis Steps
The next step is clustering analysis based on the K-Means method. This analysis
was carried out to group the data into three clusters. In this step, the data is grouped based
on three variables, namely the number of visitors, number of buyers, and total sales. The
stages carried out for this analysis include:
1. Selection of the number of clusters (K)
At this stage, the number of clusters (K) is determined. Choosing the right K is one
of the key stages in K-Means. Indicators that can be used to determine the number of
clusters are usually Elbow or Silhouette. The number of clusters in the research was
determined to be 3 to represent the low-selling, normal and best-selling sales clusters.
2. Determination of Cluster Centre
Determining the cluster centre is carried out to obtain the cluster centre based on
the Euclidean method, as in Equation (1). Determining the cluster centre uses random
numbers (arbitrary) as initialization. The initial cluster centre is the initial reference in the
clustering process.
3. Cluster Centre Iteration Based on K-Means
Iterations are carried out until convergence; that is, there are no significant changes
in the placement of data into clusters. Each iteration involves several steps, including
calculating the distance of a data point to the nearest cluster centre, determining cluster
members, calculating a new cluster centre based on the average of its members (K-
Means), and finally repeating these steps until it converges.
4. Determination of Cluster Labels
Vies Sata Zullah, Raden Mohamad Atok
Jurnal Indonesia Sosial Teknologi, Vol. 5, No. 9, September 2024 3420
After obtaining three clusters, the next step is to determine each cluster's label or
name. A best-selling cluster is defined as the cluster with the highest cluster centre value,
while a low cluster is defined as the cluster with the lowest cluster centre value.
Forecasting Steps
The experience process is carried out by applying the exponential smoothing
method. Exponential smoothing allows us to identify trends and patterns in shoe sales
data in best-selling groups. By forecasting these Ventela shoe models, companies or
sellers can better prepare for changes in demand and can optimize inventory and sales
strategies in hot groups. The following are the stages of Exponential Smoothing
forecasting in detail.
1. Selection of the Exponential Smoothing Model
Select the Exponential Smoothing (ES) model that is most appropriate for the data
used. Several types of ES models can be used, such as Single Exponential Smoothing
(SES), Double Exponential Smoothing (Holt), and Triple Exponential Smoothing (Holt-
Winters). The choice of model depends on whether the data has a trend, seasonal
component, or both. If the use of visuals does not sufficiently confirm the presence of
trends, seasonality, or both, then decomposition is required.
2. Initialize Model Parameters
Initialization of the initial model parameters determines the initial parameters to be
applied. For example, in the Single Exponential Smoothing (SES) model, it is necessary
to initialize the alpha (α) value, which is the exponential weight parameter for the latest
observations. This initial value can be selected using the grid search technique.
3. Model Prediction Calculation
At this stage, prediction calculations are carried out based on the selected model.
These calculations are needed to obtain estimated (fitted) values , which are then used to
validate the goodness of the model.
4. Measurement of Prediction Error
Prediction error measurements validate the model's ability to predict actual data.
The evaluation of prediction quality for time series data usually uses Mean Absolute
Percentage Error (MAPE).
5. Error Value Checking
After measuring the model error, the next step is to check whether the error value
meets the specified requirements. For example, MAPE is categorized as good if it is less
than 20%. If it is not appropriate, then initialize the model parameters again using grid
search.
6. Data Forecasting
Cluster Analysis and Forecasting on Local Shoe Products: Case Study for Ventela in Indonesia
Jurnal Indonesia Sosial Teknologi, Vol. 5, No. 9, September 2024 3421
The final stage is to forecast the data using the methods and parameters that are
considered the most optimal. Forecasting is intended to derive some value in the future.
The length of the data forecast will be determined later and will be adjusted to the
researcher's needs.
Results and Discussion
Descriptive Statistics of Clustering Data
The data used in this research includes four variables, namely Number of Visitors,
Number of Buyers, Total Sales, and Number of Products Sold. The first three variables,
namely the Number of Visitors, Number of Buyers, and Total Sales, are used for grouping
and the remainder, namely the Number of Products Sold, are used for forecasting.
Therefore, the descriptive statistics of this research variable are divided into two types,
namely grouping variables and forecasting variables. The results of descriptive analysis
of the variables used for grouping are presented in Table 3.
Table 3
Descriptive statistics of clustering variables
Measures of
variable data
centering
Variable
Number of
visitors (
)
Number of
buyers (
)
Total Sales
(
)
Minimum
457.00
2.00
Rp 659,481.00
Median
6,954.00
170.00
Rp 49,904,657.00
Maximum
45,697.00
1,115.00
Rp 273,968,637.00
Range
45,240.00
1,113.00
Rp 273,309,156.00
Mean
10,303.00
264.00
Rp 69,261,843.49
Standard
deviation
9,786.00
282.00
Rp 67,539,149.93
Table 3 shows that the average number of visitors who only saw the Ventela shoe
model was 10,303 people, of which only 2.56% of that number purchased Ventela shoes,
with an average total sale of each shoe model of IDR 69,261,843.49. The standard
deviation obtained from the three variables has a significant value, indicating that each
shoe model has a very varied number of visitors, number of buyers and total sales, or
there are very small or very large values for these variables. This is proven by the large
range of values, where the most visited model is Republic Low White with 45,697 people,
and the least visited is Reborn Reflective Navy with 457 people. The range of buyers is
also quite large, where the most purchased model is Ethnic Low Black Natural by 1,115
people, and the least is New Public Low Dark Brown by two people. Republic Low White
produced the largest total sales, namely IDR 273,968,637.00 and the lowest total sales,
namely New Public Low Dark Brown, amounted to IDR 659,481.00. The condition of
each variable can be clarified by visualization using a box plot to see the distribution of
the data. The visualization results of the grouping variables are shown in Figure 2.
Vies Sata Zullah, Raden Mohamad Atok
Jurnal Indonesia Sosial Teknologi, Vol. 5, No. 9, September 2024 3422
(a)
(b)
(c)
Figure 2. Box plot of clustering variables
(a) Number of visitors, (b) Number of buyers, and (c) Total sales
Figure 2 shows that there are several observations (Ventela shoe model) that are
outliers in each variable. Some of the Ventela shoe models include Republic Low White,
Public Low Cream, Ethnic Low Black Natural, Republic Low Black Natural, and Public
Low Black Natural which have high sales and can be assumed to be the Ventela shoe
models that are in the best-selling group, but this statement It is just a guess without any
basis for analysis. Therefore, to ensure that the Ventela shoe models are grouped in the
best-selling cluster, a cluster analysis was carried out based on three grouping variables
using the k-means clustering method.
Clustering Shoe Models Using K-Means
This subchapter discusses the grouping results obtained from the clustering process
using the k-means clustering method. In accordance with the principles used by k-means,
the number of clusters in the grouping has been determined to be three: the low-selling,
normal, and best-selling groups. The initial cluster centre applied in the first stage of the
grouping process is presented in Table 4.
Table 4
Initial value of cluster centre
Variable
Cluster
1
2
3
Number of
visitors
1,147.00
16,496.00
45,697.00
Number of
buyers
2.00
314.00
996.00
Cluster Analysis and Forecasting on Local Shoe Products: Case Study for Ventela in Indonesia
Jurnal Indonesia Sosial Teknologi, Vol. 5, No. 9, September 2024 3423
Total sales
659,481.00
111,804,153.00
273,968,637.00
Table 4 shows that the order of groups that analogize the lowest to highest values
in sequence are clusters 1, 2, and 3. This means that cluster 1 represents the group of
Ventela shoe models that sell less well, cluster 2 represents the group of Ventela shoe
models that sell normally, and cluster 3 represents the best-selling Ventela shoe models.
Next, the cluster center will be iterated until it converges and gets the final cluster center.
Changes that occur in the cluster center values at each iteration are presented in Table 5.
Table 5
Change in cluster center value
Iteration
Cluster
1
2
3
1
23.240.594,687
16.919.588,118
46.870.502,108
2
1.159.838,129
2.347.244,801
9.749.313,140
3
0,000
0,000
0,000
After the iteration to get the cluster center has converged, then the most optimal
cluster center value is obtained. The optimal cluster center results are shown in Table 6.
Table 6
Optimal cluster center value
Variable
Cluster
1
2
3
Jumlah
Pengunjung
4.046,21
14436,33
29.170,50
Jumlah Pembeli
81,00
364,53
869,17
Total Penjualan
25.059.913,64
92.537.320,20
217.348.824,33
As in Table 6, the final cluster center values in Table 4.4 give the same cluster
labels. The order of cluster center values from low to high is clustered 1, 2, and 3 so that
cluster 1 represents the group of Ventela shoe models that sell less well, cluster 2
represents the group of Ventela shoe models that sell normally, and cluster 3 represents
the Ventela shoe models that sell well. The number of Ventela shoe models grouped into
each cluster is shown in Table 7 below.
Table 7
Number of cluster members
Cluster
Number of Members
1 (low-selling)
28
2 (normal-selling)
15
3 (best-selling)
6
Based on Table 7, there are only 6 Ventela shoe models included in the best-selling
cluster (cluster 3). These shoe models are presented in Table 8.
Vies Sata Zullah, Raden Mohamad Atok
Jurnal Indonesia Sosial Teknologi, Vol. 5, No. 9, September 2024 3424
Table 8. Ventela shoe model in the best-selling cluster
No
Model
Color
1
Ventela Ethnic Low
All Black
2
Ventela Ethnic Low
Black Natural
3
Ventela Public Low
Black Natural
4
Ventela Public Low
Cream
5
Ventela Republic Low
Black Natural
6
Ventela Republic Low
White
Based on the Ventela shoe model contained in Table 8, these results indirectly
confirm observations that are categorized as outliers in the boxplot for the variables used
in the grouping process.
Descriptive Statistics of Forecasting Data
After obtaining members from the best-selling cluster group, the next step is to
forecast the number of product sales of the Ventela shoe models included in the cluster.
The data that will be predicted is the total products sold for each of the six Ventela shoe
models, according to Table 8. The main objective of this forecasting is to determine the
potential product sales of each model of Ventela shoes that sell well to avoid the
possibility of stockouts. An illustration of the movement in the number of shoe product
sales in the best-selling cluster is shown in Figure 3 below.
Figure 3. Illustration of the sales number of best-selling cluster shoe products
Figure 3 shows that the movement pattern of the number of products sold for the
six shoe models has a trend that tends to be irregular. At the beginning of 2022, all shoe
models experience a relatively high increase in sales until their peak in mid-2022. After
that, the graph of total products sold will decline until they show a sloping pattern and
tend to decline. The pattern of an increase followed by a decrease (up then down) shows
that there are two possible forecasting models, namely Double Exponential Smoothing
(DES) and Triple Exponential Smoothing (TES).
Cluster Analysis and Forecasting on Local Shoe Products: Case Study for Ventela in Indonesia
Jurnal Indonesia Sosial Teknologi, Vol. 5, No. 9, September 2024 3425
Data patterns can also be identified using boxplots. Boxplots are created based on
data in the same month. For example, the number of products sold in January means that
the boxplot is built from data on the number of products sold each January in all years of
observation. Boxplot data on the number of products sold is illustrated in Figure 4.
(a)
(b)
(c)
(d)
(e)
(f)
Figure 4. Monthly boxplot visualization of the number of products sold
(a) Ventela Ethnic Low All Black, (b) Ventela Ethnic Low Black Natural,
(c) Ventela Public Low Black Natural, (d) Ventela Public Low Cream,
(e) Ventela Republic Low Black Natural, dan (f) Ventela Republic Low White
The Boxplot illustrated in Figure 4 shows that the sales trend for all shoe models in
large quantities tends to occur at the end of the year, starting from July to October and
then sloping again (Putra, 2022). This information implies a seasonal trend likely to occur
in data on the number of products sold. However, a reasonably long boxplot body also
means that the alleged seasonality does not strongly influence the number of products
sold. Therefore, a good ES model to use for forecasting has yet to be determined with
Vies Sata Zullah, Raden Mohamad Atok
Jurnal Indonesia Sosial Teknologi, Vol. 5, No. 9, September 2024 3426
certainty, and modeling simulations are needed to try both possible ES models that can
be used.
Data Forecasting Using Exponential Smoothing
The data patterns for the six shoe models tend to be irregular, resulting in no direct
determination regarding the most suitable Exponential Smoothing (ES) model. Therefore,
experiments are carried out in the forecasting process to apply the possible types of ES
models, namely DES and TES. The simulated TES model is additive, because the
seasonal pattern does not show any multiplication (increase over time). The absence of
provisions regarding the selection of weight values in ES modeling makes it necessary to
determine which weights represent small (0.25), medium (0.50), and large (0.75). The
entire model is built by adjusting the three predetermined weights.
Double Exponential Smoothing Modelling
The next data modeling was done using the Double Exponential Smoothing (DES)
method. This model involves two weights, namely α and β, which are coefficients or
smoothing weights for level and trend. The values α and β are determined according to
the previous provisions so that nine different models are formed for each shoe model. The
best model for the six shoe models in the best-selling cluster is presented in Table 9.
Table 9
Best DES model for each shoe model
No
Shoe model
MAPE (%)
Model equation
1
Ventela Ethnic
Low All Black
162.25
2
Ventela Ethnic
Low Black Natural
49.48
3
Ventela Public
Low Black Natural
63.21
Cluster Analysis and Forecasting on Local Shoe Products: Case Study for Ventela in Indonesia
Jurnal Indonesia Sosial Teknologi, Vol. 5, No. 9, September 2024 3427
No
Shoe model
MAPE (%)
Model equation
4
Ventela Public
Low Cream
110.96
5
Ventela Republic
Low Black Natural
78.99
6
Ventela Republic
Low White
69.53
Based on Table 9, the DES model MAPE value for all shoe models is large, greater
than 20%. This is thought to be caused by irregular data patterns. Even though the
complexity of the DES model has increased with the Parameter compared to the SES
model, it is still unable to capture the data patterns we want to predict well. Therefore,
these models cannot be considered good models for predicting data.
Triple Exponential Smoothing Modelling
The final data modeling used the Triple Exponential Smoothing (TES) method.
This model is the most complex of the other ES models, which involves three weights,
namely. , , and , Each of which is a smoothing coefficient or weight at level, trend,
and seasonality. The values α, β, and γ are determined according to the previous
provisions so that 27 different models are formed for each shoe model. The best model
for the six shoe models in the best-selling cluster is presented in Table 10.
Table 10
Best TES model for each shoe model
No
Shoe model
MAPE
(%)
Model equation
1
Ventela Ethnic
Low All Black
133.73
Vies Sata Zullah, Raden Mohamad Atok
Jurnal Indonesia Sosial Teknologi, Vol. 5, No. 9, September 2024 3428
No
Shoe model
MAPE
(%)
Model equation
2
Ventela Ethnic
Low Black
Natural
29.17
3
Ventela Public
Low Black
Natural
53.08
4
Ventela Public
Low Cream
119.88
5
Ventela
Republic Low
Black Natural
138.26
6
Ventela
Republic Low
White
96.62
Table 10 shows that the MAPE value for the TES model is still not much different
from the DES model, where for all shoe models, it has a large value, namely greater than
20%. In some shoe models, the complexity of the TES model worsens the prediction
results obtained. Therefore, these models cannot be considered as good models for
predicting data.
Selection of the Best Model Based on MAPE Accuracy
Cluster Analysis and Forecasting on Local Shoe Products: Case Study for Ventela in Indonesia
Jurnal Indonesia Sosial Teknologi, Vol. 5, No. 9, September 2024 3429
After modeling using two ES models, the best model is selected globally, namely
the best model from the best models in each model (DES and TES). Based on the MAPE
obtained in the three models, the best model globally is obtained, as presented in Table
11.
Table 11
Best ES model for each shoe model
No
Shoe model
MAPE
(%)
Model equation
1
Ventela
Ethnic Low
All Black
TES
(0,5; 0,5;
0,25)
2
Ventela
Ethnic Low
Black
Natural
TES
(0,5; 0,5;
0,25)
3
Ventela
Public Low
Black
Natural
TES
(0,75;
0,25;
0,25)
4
Ventela
Public Low
Cream
DES
(0,25;
0,75)
5
Ventela
Republic
Low Black
Natural
DES
(0,25;
0,75)
6
Ventela
Republic
Low White
DES
(0,25;
0,75)
The MAPE value obtained from the best model in Table 11 still has a large value
(>20%), but the best model must still be selected for forecasting. Figure 5 below
illustrates the prediction results obtained based on the best model for each shoe model.
Vies Sata Zullah, Raden Mohamad Atok
Jurnal Indonesia Sosial Teknologi, Vol. 5, No. 9, September 2024 3430
(a)
(b)
(c)
(d)
(e)
(f)
Figure 5 Visualization of Actual Comparison and Best ES Model Predictions
(a) Ventela Ethnic Low All Black, (b) Ventela Ethnic Low Black Natural, (c) Ventela Public
Low Black Natural, (d) Ventela Public Low Cream, (e) Ventela Republic Low Black Natural,
dan (f) Ventela Republic Low White
Figure 5. shows that the model predictions obtained for each shoe model follow the
actual data pattern. Thus, visually, the model can be considered capable of predicting the
model well and can be used to predict the number of products for each shoe model.
Forecasting Results and Illustrations
Based on the best model explained in the previous subsection, forecasting is carried
out to obtain a projected number of product sales for each shoe in the next few months.
In this study, the number of predictions was determined to be 12, which refers to the
Cluster Analysis and Forecasting on Local Shoe Products: Case Study for Ventela in Indonesia
Jurnal Indonesia Sosial Teknologi, Vol. 5, No. 9, September 2024 3431
frequency of data used, namely monthly. Forecasting results for each shoe model are
presented in Table 12.
Table 12
Forecasting result for each shoe model
Periode
V.
Ethnic
Low All
Black
V. Ethnic
Low
Black
Natural
V. Public
Low
Black
Natural
V.
Public
Low
Cream
V.
Republic
Low
Black
Natural
V.
Republic
Low
White
April 2024
10
20
20
3
0
1
Mei 2024
30
36
27
1
0
0
Juni 2024
19
32
19
0
0
0
Juli 2024
106
44
34
0
0
0
Agustus
2024
20
34
28
0
0
0
September
2024
23
35
17
0
0
0
Oktober
2024
15
32
17
0
0
0
November
2024
28
18
8
0
0
0
Desember
2024
23
36
21
0
0
0
Januari
2025
30
26
3
0
0
0
Februari
2025
14
10
6
0
0
0
Maret 2025
13
18
1
0
0
0
* V. stands for Ventela
To clarify the forecast movement pattern obtained, an illustration is carried out as
in Figure 6.
(a)
(b)
Vies Sata Zullah, Raden Mohamad Atok
Jurnal Indonesia Sosial Teknologi, Vol. 5, No. 9, September 2024 3432
(c)
(d)
(e)
(f)
Figure 6. Illustration of the best es model forecast results
Ventela Ethnic Low All Black, (b) Ventela Ethnic Low Black Natural, (c) Ventela Public
Low Black Natural, (d) Ventela Public Low Cream, (e) Ventela Republic Low Black
Natural, dan (f) Ventela Republic Low White
Based on Figure 6, it can be seen that there are forecast results that show fluctuating
value movements for forecasts using the TES model. However, forecasts using the DES
model show a constant decreasing trend. However, the constant forecast results can still
be used as a reference because they have a range of forecast intervals or Prediction
Intervals (PI) that could occur. Therefore, to maximize the sales potential of each shoe
model, sellers must prepare a stock of the maximum value of the PI range obtained so
that stockouts do not occur.
Managerial Implications
The managerial implications of this research relate to the forecasting results for
each shoe model in the best-selling cluster. Several shoe models, such as Public Low
Cream, Republic Low Black Natural, and Republic Low white, show no sales for the next
few months or have a forecast result 0. A shoe model with a forecast result of 0 means
that the model no longer has sales power or has not been sought after by buyers. Sellers
can consider not re-stocking this model or only having a maximum of 1 stock as an
opportunity to make a profit. Shoe models that previously had high sales will not
necessarily have high sales in the future.
Cluster Analysis and Forecasting on Local Shoe Products: Case Study for Ventela in Indonesia
Jurnal Indonesia Sosial Teknologi, Vol. 5, No. 9, September 2024 3433
Sellers can cut the selling price of shoes to reduce the stock in the warehouse or
shelves so that they do not only sit for a short time because selling power has decreased.
The sales proceeds can be used as capital to replace it with the latest model. Sellers may
also consider adding other brands while Ventela produces new models. This can fill
empty warehouses or shelves by replacing models that have expired so that sellers still
have the opportunity to make a profit. The shelf life of each shoe model is different.
Therefore, sellers can estimate the shelf life from forecast results to avoid overstock.
Conclusion
Based on the results of grouping using the K-Means clustering method on shoe
models with three variables (number of visitors, number of buyers, and total sales), it was
found that the number of shoe models included in the best-selling cluster was six. The six
shoe models include (1) Ventela Ethnic Low All Black, (2) Ventela Ethnic Low Black
Natural, (3) Ventela Public Low Black Natural, (4) Ventela Public Low Cream, (5)
Ventela Republic Low Black Natural, and (6) Ventela Republic Low White.
The ES model's accuracy value (MAPE) in forecasting obtained is above 20%,
which shows that the Exponential Smoothing model cannot model the number of shoe
products sold for each shoe model well. Referring to the sample of this study, which only
spans less than three years, several shoe models produce a forecast value of zero
(especially for Ventela Public Low Cream, Ventela Republic Low Black Natural, and
Ventela Republic Low White). Based on the forecast results for the next 12 months, the
three shoe models included in the best-selling cluster will have no sales.
Bibliography
Alwiyah, M., & Gata, G. (2019). E-Commerce Untuk Meningkatkan Penjualan Sepatu
Sneakers Bintaro. IDEALIS: InDonEsiA JournaL Information System, 2(6), 345–
350.
Dwiarni, B. A., & Setiyono, B. (2020). Akuisisi dan Clustering Data Sosial Media
Menggunakan Algoritma K-Means sebagai Dasar untuk Mengetahui Profil
Pengguna. Jurnal Sains Dan Seni ITS, 8(2), A65–A70.
Jananto, A. (2022). Perbandingan Analisis Cluster Algoritma K-Means Dan AHC Dalam
Perencanaan Persediaan Barang Pada Perusahaan Manufaktur. Progresif: Jurnal
Ilmiah Komputer, 18(2), 257–270.
Maricar, M. A. (2019). Analisa perbandingan nilai akurasi moving average dan
exponential smoothing untuk sistem peramalan pendapatan pada perusahaan xyz.
Jurnal Sistem Dan Informatika (JSI), 13(2), 36–45.
Pérez-Ortega, J., Almanza-Ortega, N. N., Vega-Villalobos, A., Pazos-Rangel, R., Zavala-
Díaz, C., & Martínez-Rebollar, A. (2019). The K-means algorithm evolution.
Introduction to Data Science and Machine Learning, 69–90.
Vies Sata Zullah, Raden Mohamad Atok
Jurnal Indonesia Sosial Teknologi, Vol. 5, No. 9, September 2024 3434
Pratiwi, W. A., & Marizal, M. (2022). Penerapan Metode Eksponential Smoothing Dalam
Memprediksi Hasil Pencapaian Kinerja Pelayanan Perangkat Daerah Dinas
Pendidikan Provinsi Riau. Indonesian Council of Premier Statistical Science, 1(1),
4–14.
Putra, Y. Y. (2022). Perlindungan Hukum Pemegang Hak Merek Deenay Sebagai Merek
Terkenal Dari Produk Yang Dipasarkan Melalui Platform E-Commerce
Berdasarkan Uu No. 20 Tahun 2016 Tentang Merek Dan Indikasi Geografis Juncto
Uu No 19 Tahun 2016 Tentang Informasi Dan Transaksi Elektronik. Fakultas
Hukum Universitas Pasundan.
Rachmawati, N. L., & Lentari, M. (2022). Penerapan metode Min-Max untuk Minimasi
Stockout dan Overstock persediaan bahan baku. Jurnal INTECH Teknik Industri
Universitas Serang Raya, 8(2), 143–148.
Shi, C., Wei, B., Wei, S., Wang, W., Liu, H., & Liu, J. (2021). A quantitative discriminant
method of elbow point for the optimal number of clusters in clustering algorithm.
EURASIP Journal on Wireless Communications and Networking, 2021, 1–16.
Yusditara, W., Damanik, H. M., Purba, M. L., & Samosir, H. E. S. (2022). Pengaruh
Brand Image Dan Promosi Terhadap Keputusan Pembelian Sepatu Lokal Merek
Compass. Jurnal Ilmiah Simantek, 6(2), 27–32.
Zullah, V. S. (2024). Analisis Klaster dan Peramalan Pada Produk Sepatu Lokal. Institut
Teknologi Sepuluh Nopember.
Zunaidi, R. A. (2022). Market Simulator Konsep Produk Baru Sepatu Wanita Bagi
UMKM Klastik Footwear. Journal of Advances in Information and Industrial
Technology, 4(1), 29–36.
