Zeta Ae Math Journal Volume 10 No.
May 2025, pp.
E-ISSN: 2579-5864 P-ISSN: 2459-9948
D https://doi.
org/10.
31102/zeta.
Application of Support Vector Regression in Time Series Analysis of Dior Stock Prices Adma Novita Sari1.
Talitha Zuleika1.
Fariz Fadillah Mardianto1*.
Elly Pusporani1 Statistics Study Program.
Faculty of Science and Technology.
Universitas Airlangga.
Indonesia *Corresponding Author Email: m.
m@fst.
ABSTRACT
Christian Dior (Dio.
is a multinational company focusing on luxury goods, including fashion products, cosmetics, and accessories.
In 2020Ae2024.
Dior's share price will experience significant fluctuations influenced by financial performance, global market trends, etc.
These fluctuations require investors to implement appropriate strategies to minimize the risk of losses and support sustainable economic growth.
This step aligns with goal 8 of the Sustainable Development Goals (SDG.
, emphasizing the importance of sustainable economic growth through investment and infrastructure development for economic prosperity.
One of the effective methods for modeling and predicting stock prices is Support Vector Regression (SVR).
By applying SVR using the Radial Basis Function (RBF) kernel, this study shows that the model can generate predictions with a MAPE value of 2.
5864% on the test data.
The SVR method is expected to provide accurate predictions, making it a helpful tool for investors and market analysts to make better investment decisions.
Keyword: DiorAos Stock Price.
Radial Basis Function.
Support Vector Regression Article info:
Submitted: December 21, 2025 Accepted: May 28, 2025 How to cite this article:
Sari.
Zuleika.
Mardianto.
F, & Pusporani.
Application of Support Vector Regression in Time Series Analysis of Dior Stock Prices.
Zeta - Math Journal, 10.
, 51-60.
https://doi.
org/10.
31102/zeta.
This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution-ShareAlike 4.
0 International License.
Adma Novita Sari, dkk.
Application of Support Vector Regression in Time Series Analysis of Dior Stock Prices INTRODUCTION Stock price movements are one of the main indicators reflecting a companyAos performance as well as the overall market condition.
Stock price fluctuations provide an overview of how a company is perceived by investors based on financial performance, future prospects, and responses to global economic dynamics.
As a critical benchmark, stock prices also reflect the level of investor confidence in the stability and growth of a company in facing market challenges (Ngui, 2.
In addition, changes in stock prices can serve as a benchmark for investors in terms of making investments to enhance company value (Pratama et al.
, 2.
Amid increasingly complex market dynamics, observing stock price fluctuations becomes crucial for investors and market analysts to understand market trends and predict changes that might affect asset values.
This aligns with the goals of the Sustainable Development Goals (SDG.
, particularly Goal 8, which focuses on sustainable economic growth through smart investment (United Nations, 2.
Christian Dior, or more commonly known as Dior, is a multinational company from France that operates in the luxury goods industry.
Founded in 1946.
Dior has become one of the most renowned brands in the world, offering a range of products including clothing, cosmetics, perfumes, jewelry, and accessories.
The company is known for its elegant and innovative designs, which continue to influence the global fashion industry.
As a company in the luxury goods sector.
DiorAos stock performance is heavily influenced by global trends in luxury consumption, macroeconomic conditions, and market dynamics.
During the period 2020Ae2024.
DiorAos stock prices showed significant fluctuations due to various factors, including the impact of the COVID-19 pandemic, changes in consumer behavior, and global economic recovery.
Sustained demand for luxury goods, especially in Asian markets, has helped support DiorAos stock value, with major contributions from the cosmetics and fashion segments (Ningrum & Sukresna, 2.
Various methods have been developed to analyze and predict stock prices.
One popular approach is Support Vector Regression (SVR), which is a modification of Support Vector Machine (SVM) for regression A previous study on stock price forecasting for ADRO using SVR resulted in a MAPE value of 45% (Maghfirah et al.
, 2.
Additionally, another study predicted gold prices with SVR using a grid search algorithm and achieved a MAPE value of 3.
73% (Gunanta et al.
, 2.
From these studies, most have not specifically applied SVR to luxury goods industry stocks such as Dior using the Radial Basis Function (RBF).
RBF is one of the kernel functions commonly used in SVM and SVR methods.
This function helps the model handle data with nonlinear patterns by transforming the data so that complex relationships can be better recognized (GeeksforGeeks, 2.
In recent research, more complex machine learning-based methods, such as deep learning, have also been used for stock price prediction.
However, these methods tend to require large amounts of data and longer computational times compared to SVR (Faisol et al.
, 2.
Therefore.
SVR remains a relevant and practical method for stock price analysis, especially when available data is limited.
The scientific novelty of this research lies in the development of a Dior stock price prediction model using Support Vector Regression (SVR) with a Radial Basis Function (RBF) kernel designed to capture nonlinear stock price fluctuation patterns during the period 2020Ae2024.
Unlike previous studies that predominantly focused on banking stocks or stock indices such as the IDX Composite (IHSG), this study introduces the context of the luxury goods industry, particularly Dior, which has unique market characteristics and dynamics.
The hypothesis of this research is that SVR with an RBF kernel can produce a reliable prediction model with a low error rate measured by the Mean Absolute Percentage Error (MAPE).
The aim of this study is to develop an accurate Dior stock price prediction model based on SVR, which can serve as a decision-support tool for investors and market analysts in making better investment decisions.
This research is expected to provide significant contributions in the field of stock market analysis, particularly in the luxury goods industry.
LITERATURE REVIEW
1 Dior Stock Price Christian Dior (Dio.
is a multinational company operating in the luxury goods industry, offering products that include fashion, cosmetics, and accessories.
DiorAos stock is traded on the Paris Stock Exchange.
In recent years.
DiorAos stock price has exhibited significant fluctuations, influenced by various factors such as the Zeta Ae Math Journal.
Vol.
No.
1, pp.
51 - 60.
May, 2025.
companyAos financial performance, global market trends, and dynamics within the luxury goods industry.
According to data from Yahoo Finance.
DiorAos stock price reached a 52-week high of C832.
50 on March 14, 2024, and a 52-week low of C529.
50 on November 12, 2024 (Yahoo Finance, 2.
These fluctuations reflect the stockAos sensitivity to market conditions and other external factors.
Analyzing DiorAos stock price is crucial to understanding market behavior and assisting investors in making informed decisions.
2 Support Vector Regression (SVR) Support Vector Regression (SVR) is an application of the Support Vector Machine (SVM) method designed for regression tasks.
One of the methods used to model and predict stock prices is Support Vector Regression (SVR).
SVR has been applied in various studies to estimate stock price movements and has proven effective in identifying patterns in time series data and solving nonlinear problems (Saadah, 2.
SVR
produces outputs in the form of real or continuous values.
The main advantage of this method lies in its ability to handle overfitting issues, thus achieving optimal performance.
SVR models are often utilized to minimize the Mean Square Error (MSE).
The SVR algorithm is an advancement in machine learning theory and possesses strong capabilities in solving prediction accuracy problems (Arfan & Lusiana, 2.
The goal of the SVR algorithm is to determine the optimal hyperplane as the best separating line.
Suppose there is a training dataset {.
cu1 , yc1 ), .
cu2 , yc2 ).
A , .
cuycn , ycycn )}, ycn = 1,2.
A , ycc when ycuycn OO Eyycc , with ycc is a dimension and ycycn is a result score (Purwoko et al.
, 2.
The following is the SVR model equation.
= yeoycN yuc.
yeo is an n-dimensional weight vector yuc.
is a function in n-dimensional space that maps ycu yca is the bias 3 Kernel Function in SVR The performance of the SVR method heavily depends on the choice of kernel function used.
The selected kernel function can be applied as a substitute to help overcome issues of data nonlinearity in high-dimensional spaces (Liu et al.
, 2.
A kernel function is dot product yuc.
cuycn ).
which can be formulated into the following equation .
cuycn , yc.
= yuc.
cuycn ).
Thus, the SVR regression function can be explained as follows.
ycycn = yce.
cuycn ) = Oc.
caycn Oe ycaycnO ) ya.
eoycn , yc.
ycn=1 Parameters and kernel functions in the SVR model need to be properly tuned because they can affect the accuracy level in making predictions.
Table 1 presents several kernel functions that are commonly used in SVR Table 1.
Kernel Functions in SVR Types of Kernel Functions Formula Linier ya.
cuycn , yc.
= ycuycnycN ycu Polynomial ya.
cuycn , yc.
= .
cuycnycN yc.
ycy , ycy = 1,2.
A Radial Basis Function (RBF) ya.
cuycn , yc.
= exp (Oey.
cuycn Oe yc.
| ) Sigmoid ya.
cuycn , yc.
= ycycaycuEa .
uycuycnycN , ycu y.
Source: Isnaeni et al.
, 2022
ycycaycuEa is a hyperbolic function that produces values between -1 and 1 yci is a kernel parameter that determines how sensitive the kernel is to differences in the input.
ycN is the transpose operator that produces the transpose of a matrix.
Adma Novita Sari, dkk.
Application of Support Vector Regression in Time Series Analysis of Dior Stock Prices 4 Grid Search Optimization Grid search optimization is a technique used to find the optimal combination of parameters for a model by systematically evaluating all possible parameter combinations that have been predefined (Fajri & Primajaya, 2.
To search for the optimal parameters, the grid search method divides the range of parameters to be optimized and iterates through each point.
The best SVR cross-validation parameters are determined by the grid search algorithm.
The main goal is to find the best combination of hyperparameters that can predict the test data with high accuracy.
The grid search method for finding the optimal hyperparameters requires a considerable amount of time.
Therefore, the grid-based hyperparameter search process is carried out in two stages: loose grid and finer grid (Purnama & Hendarsin, 2.
The loose grid stage is the initial step where the optimal parameters are selected from integer power values.
Next, the finer grid stage is conducted by searching for the optimal parameters around the values obtained in the loose grid stage to achieve more precise results.
5 Mean Absolute Persentage Error (MAPE) One of the criteria for determining the goodness of prediction results from the model is by looking at the Mean Absolute Percentage Error (MAPE).
MAPE shows the level of absolute error of the prediction results.
The MAPE value can be obtained using the following formula.
Ocycu1=1 ycAyaycEya = ycycn is the observed value at time period t ycCycn is the predicted value at time period t ycu is the number of prediction data points .
cycn Oe ycCycn | ycycn y 100% ycu The magnitude of the MAPE value can be interpreted as the level of accuracy of the predictions produced by a model.
A model that produces a MAPE value below 10% indicates that the model yields accurate predictions, while a MAPE value above 50% indicates that the model produces inaccurate predictions.
The interpretation of the MAPE values can be seen in Table 2.
Table 2.
MAPE Value Criteria
MAPE
Accuracy Rate
MAPE < 10%
The prediction results are classified as very accurate 10% O MAPE < 20% The prediction results are relatively accurate 20% O MAPE < 50% The prediction results are considered worthy MAPE Ou 50% The predicted results are relatively poor Source: Moreno et al.
, 2013
METHOD
1 Data Source and Variables This research was conducted using a quantitative approach focusing on time series data analysis.
The analyzed data comprises DiorAos stock prices obtained from the investing.
com website.
The study utilizes weekly data covering the period from August 2020 to November 2024.
The research data is divided into two parts, namely training data and testing data, with 90% allocated for training and 10% for testing.
The training data will be used to build the model, covering data from the first week of August 2020 to the last week of May Meanwhile, the testing data will be used to assess the model's accuracy, covering data from the first week of June 2024 to the last week of November 2024.
The variable in this study is DiorAos stock price.
2 Stages of Data Analysis Data analysis is conducted using R software.
The following outlines the detailed stages of data analysis applied in this study.
Zeta Ae Math Journal.
Vol.
No.
1, pp.
51 - 60.
May, 2025.
Determine the characteristics of DiorAos stock price by creating a time series plot and performing descriptive statistical analysis.
Split the research data into two parts: training data and testing data, with proportions of 90% and 10%.
Conduct Terasvirta test and White test to examine the assumptions of linearity and heteroscedasticity in the data.
Generate a PACF plot to determine the significant lags in DiorAos stock price data.
Perform initial modeling to determine the best kernel function for predicting DiorAos stock price data based on the criteria of minimum RMSE and MAPE values.
Tune parameters using the grid search method in two stages: loose grid and finer grid.
Model SVR on the training data using the optimal parameters obtained from the grid search tuning process.
Predict DiorAos stock prices on the testing data based on the best SVR model.
Calculate the MAPE value from the predictions on the testing data and draw conclusions.
The series of analytical stages described earlier can be more clearly and systematically illustrated through the following flowchart Determining the characteristics of Dior's stock price data Division of training data and testing data Linearity and heteroskedasticity test Forming a PACF plot of Dior's stock price Selection of the first model of the SVR method with the best kernel function Tuning parameters Modeling of SVR on optimal parametric training data Prediction of test data with best SVR model Calculate MAPE values and draw Figure 1.
Flowchart of Data Analysis Stages (Source: Analysis, 2.
RESULT AND DISCUSSION 1 Characteristics of Dior Share Price Data In this study, the characteristics of Dior's stock prices are visualized using a time series plot and explained through descriptive statistics, which include the number of observations, mean, minimum, and maximum Prior to this, the Dior stock price data is divided into training and testing sets with a respective proportion of 90% and 10%.
According to Aisyah et al.
, the proportion for splitting training and testing data is subjectively determined by the researcher.
The time series plot of Dior stock prices is presented below.
Figure 2.
Time Series Plot (Source: Analysis, 2.
Adma Novita Sari, dkk.
Application of Support Vector Regression in Time Series Analysis of Dior Stock Prices Based on Figure 2, the difference in line colors on the plot is used to distinguish between training and testing data.
It can be identified that Dior's stock prices exhibit an upward trend from 2020 to 2024, with weekly fluctuations.
The highest stock price appears on the 141st data point, dated April 9, 2023, while the lowest is on the 1st data point, dated August 2, 2020.
The results of the descriptive statistical analysis of Dior stock prices are presented in detail in Table 3.
Table 3.
Descriptive Statistics of Dior Stock Price Data Data Amount of Data Mean Minimum Maksimum Overall Data Training Data Testing Data Source: Analysis, 2024 Based on Table 3, using the 90% and 10% data split, a total of 199 training data points are obtained with the lowest stock price of 340.
4 EUR and the highest of 866 EUR.
Additionally, 27 testing data points are identified, with respective minimum and maximum stock prices of 548.
5 EUR and 616.
44 EUR.
A total of 226 observations will be used in the time series analysis using the SVR method.
2 Test of Linearity and Heteroscedasticity of Dior Share Price Data To verify the presence of nonlinearity and heteroskedasticity in Dior stock price data, tests such as the Terasvirta test and White test are employed.
The Terasvirta test is used to assess linearity in the data, and the following hypothesis is applied ya0 : The data contains a linear pattern ya1 : The data does not contain a linear pattern Subsequently, the White test is performed to detect heteroskedasticity in the data, using the following ya0 : Data are not indicated heteroscedasticity ya1 : Data indicated heteroscedasticity Using the testing criterion to reject ya0 if the ycy Oe ycycaycoycyce < yu = 5%, the results of the Terasvirta and White tests are shown in Table 4.
Table 4.
Results of Linearity and Heteroscedasticity Test
Test
P-Value Terasvirta Test
0,001143
White Test
0,004557
Source: Analysis, 2024 Based on Table 4, the p-value from the Terasvirta and White tests are 0.
001143 and 0.
004557, respectively.
With yu = 5%, the decision is to reject ya0 and it can be concluded that the data exhibits nonlinearity and Therefore, the data fulfills the assumptions required for analysis using the SVR method.
3 Dior Share Price Time Series Analysis with SVR Method The SVR method is considered a modern, non-parametric approach.
Before model fitting, the data is transformed into time lag format by identifying and removing insignificant lags with the help of a Partial Autocorrelation Function (PACF) plot.
Significant lags identified in the PACF plot are used as predictor variables (Bawues et al.
, 2.
The PACF plot for Dior's stock price is visualized in Figure 3.
Zeta Ae Math Journal.
Vol.
No.
1, pp.
51 - 60.
May, 2025.
Figure 3.
PACF Plot Dior Share Price Data (Source: Analysis, 2.
Based on Figure 3, lag 1 is identified as a significant lag because it exceeds the upper and lower bounds of the PACF plot.
Thus, lag 1 is used as the time lag input for SVR modeling.
Next, a comparison of Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE) values is performed on the initial SVR models using three kernel functions: Radial Basis Function (RBF), sigmoid, and polynomial.
The results of the initial SVR modeling on the training data are presented in Table 5.
Table 5.
Comparison of Kernel Data Training Functions Kernel
RMSE
MAPE
RBF
Sigmoid Polynomial Source: Analysis, 2024 Based on Table 5, the RBF kernel yields the best performance with an RMSE value of 24.
29531 and a MAPE value of 2.
These values indicate relatively low and stable prediction errors.
In contrast, the sigmoid kernel produces nearly three times the error of RBF, and the polynomial kernel demonstrates very poor performance with an RMSE exceeding 1000 and a MAPE above 150%, indicating a total mismatch between the polynomial kernel and the data characteristics.
This finding supports the selection of the RBF kernel as the basis for parameter tuning due to its effectiveness in capturing nonlinear patterns.
Following the initial modeling using the best-performing kernel, the SVR model's performance is enhanced through parameter tuning.
The tuning process uses a grid search method, consisting of two stages:
loose grid and finer grid search.
Each kernel function has different parameters for tuning.
For the RBF kernel, three parameters are tuned: cost .
, gamma .
, and epsilon .
uA).
The parameter value ranges used in the tuning process are presented in Table 6.
Table 6.
Grid Search Method Parameter Value Range Grid Search Method Parameter Value Range Oe5 Loose Grid Cost .
2Oe4 .
2Oe3 .
A .
Gamma .
0,1.
0,2.
A .
0,9.
Epsilon .
uA) 2Oe6 .
2Oe5 .
2Oe4 .
A .
Finer Grid Cost .
2Oe1 .
2Oe0.
2Oe0.
A .
Gamma .
2Oe1 .
2Oe0.
2Oe0.
A .
Epsilon .
uA) Source: Analysis, 2024 From the parameter ranges listed in Table 6, various combinations are tested to find the optimal parameter The optimal parameters from each grid search stage are summarized in Table 7.
Adma Novita Sari, dkk.
Application of Support Vector Regression in Time Series Analysis of Dior Stock Prices Table 7.
Optimal Parameters of the Grid Search Method Optimal Parameter Combination Grid Search Method Cost .
Gamma .
Epsilon .
Loose Grid Finer Grid Source: Analysis, 2024 According to Table 7, the optimal parameters obtained from the finer grid method are cost .
= 1.
= 3.
8, and epsilon .
uA) = 0.
This tuning process results in an RMSE value of 23.
49748 and a MAPE value of 2.
7184% on the training data.
Comparing the MAPE values before and after parameter tuning shows that the tuned SVR model achieves the lowest MAPE, indicating improved prediction performance.
Therefore, it can be concluded that the tuning process enhances the SVR model's accuracy in forecasting Dior's stock With the initial modeling completed, optimal parameters obtained, and model tuning performed, prediction of Dior's stock prices on the testing data using the SVR method can be conducted.
4 Calculation of MAPE Value The next step involves calculating the Mean Absolute Percentage Error (MAPE) to measure the prediction The calculation is carried out using the predicted stock prices of Dior during the 2020Ae2024 period against the testing dataset.
The MAPE calculation results are presented in Table 8.
Table 8.
Calculation of MAPE Value Prediction Data Testing Results (EUR) Using the RBF Kernel
Period
Testing Data Training Data Difference APE
MAPE
26/05/2024
02/06/2024
09/06/2024
16/06/2024
23/06/2024
30/06/2024
07/07/2024
14/07/2024
21/07/2024
28/07/2024
04/08/2024
11/08/2024
18/08/2024
25/08/2024
01/09/2024
08/09/2024
15/09/2024
22/09/2024
29/09/2024
06/10/2024
13/10/2024
20/10/2024
27/10/2024
03/11/2024
10/11/2024
17/11/2024
24/11/2024
Source: Analysis, 2024 As shown in Table 8, the MAPE value obtained from the predicted Dior stock price data using the SVR method with the Radial Basis Function (RBF) kernel is 2.
Thus, it can be concluded that the SVR method is highly effective and accurate in forecasting DiorAos stock price over the 2020Ae2024 period, as the resulting MAPE value is below 10%.
This finding is in line with the study conducted by Muhammad et al.
, which Zeta Ae Math Journal.
Vol.
No.
1, pp.
51 - 60.
May, 2025.
states that a MAPE value of less than 10% indicates a highly accurate prediction.
The predicted results on the testing dataset are visualized in the chart shown in Figure 4.
Figure 4.
SVR Method Data Testing Prediction Results (Source: Analysis, 2.
CONCLUSION Based on the results of this study, it can be observed that the stock price of Dior fluctuated on a weekly basis during the 2020 - 2024 period, as shown by the time series plot.
Using the SVR approach, the general model form applied to predict Dior's stock price is the one using the RBF kernel.
This model produced a MAPE value of 2.
586% on the testing dataset.
Further research is recommended to evaluate the prediction of DiorAos stock price using alternative forecasting methods, such as machine learning or deep learning, which may offer a more accurate approach in analyzing stock price patterns.
Understanding and applying appropriate forecasting methods can provide investors with better insights in managing their investments.
It is also important to note that stock price prediction cannot fully eliminate risk.
By acknowledging limitations and updating analyses regularly, economic actors can make more informed and adaptive decisions when managing their investments in DiorAos stock.
ACKNOWLEDGEMENT
The author sincerely thanks the data provider.
Investing.
com, for their valuable contribution to the implementation of this study.
Special thanks are also extended to the Undergraduate Statistics Program.
Faculty of Science and Technology.
Universitas Airlangga, as well as all parties who have supported and contributed to the completion of this research.
REFERENCES