FORECASTING WITH WEIGHTED MOVING AVERAGE METHOD FOR PRODUCT PROCUREMENT STOCK

Dhanty Store is a family start-up located in East Jakarta. It was initiated in 2018, engaged in retail with the main product in the form of women's clothing and accessories. One of the important processes in Dhanty Store operations is the product procurement process. Currently, Dhanty Store request products according to their wishes without looking at their sales data. This causes their product stock is not well controlled. When there is a lot of demand, sometimes Dhanty Shops run out of stock so their customers will move to other stores. In addition, the process of requesting and procuring products to suppliers also takes a long time so that it can further disrupt the operations of Dhanty Store. This study develops a forecasting application prototype with the Weighted Moving Average method to assist Dhanty Store in the process of requesting and procuring their products. Forecasting results in the period (t) of the 1st week of January were 275 products. In addition, this study predicts product stock with a 4-week moving average and the MAD tracking signal value is ranged from -1.51 to 3.86 and the MAPE value is 35.4%. As for the reliability and level of user acceptance of the prototype model in this study, tested using the System Usability Scale (SUS) method and it is known that the average value given by respondents was 82 with details 0% considered inappropriate, 40% considered neutral and 60% rated it according to need.


I. INTRODUCTION
Dhanty Store is a family start-up located in East Jakarta. The store was started in 2018 and is a business that is engaged in retail with its main products in the form of women's clothing and accessories. To increase its expansion, Dhanty Store also utilizes online media to market its products. Various online platforms are used such as Whatsapp, Instagram to national e-commerce to reach customers throughout Indonesia.
Based on an interview with the owner of the Dhanti Store [1], information was obtained that every day on average there are dozens of transactions for various types of products. The store continues to develop operational processes to reach customers throughout Indonesia. Currently, Dhanty Store has at least 15 employees consisting of warehouse and procurement, finance and customer service departments. The high demand for their products, especially online transactions, is not only due to the good quality but also the speed and ease of transactions. Dhanty Store strives to provide the best service to its customers.
One of the special concerns for Dhanty Store is the availability of product stock. They periodically procure products to anticipate stock outs. However, out of stock products still often occur so that their customer service must provide detailed explanations to their customers. This causes the operational activities of Dhanty Store can be disrupted and even causes their customers to move to other stores that have similar products.
Furthermore, [1] explains that the process required to request and procure a product is about 3-5 days. The process starts from requests to delivery of products from various suppliers. This will further aggravate the situation, because when the stock is empty, it can take several days for the product to be available. Moreover, currently the product request process is as desired, not based on data. Sometimes when they order a product, it doesn't meet the customer's demand.
Based on these problems, it is necessary to apply a forecasting system that can help Dhanty Store in the process of requesting and procuring their products. This forecast can be used to predict the number of products that need to be ordered at a time so that stock is continuously available and can meet customer demand. This is also to anticipate the length of the process of requesting and shipping their products. Thus, operational activities and transactions at Dhanty Store are not disrupted, even more organized and scalable.
This study developed a prototype application for Procurement Stock Forecasting with the Weighted Moving Average Method. As explained earlier that the main product from Dhanty Store is clothing which is a basic need and is used by everyday people, so the Weighted Moving Average method, which is widely used to forecast trading commodities with fairly stable data, although slightly fluctuating, is very suitable in this research [2]. In addition, this method is easier to implement when compared to other methods such as trendlines [3].
Several previous studies related to stock forecasting using various methods such as the Weighted Moving Average method by Sundari, Susanto and Revianti [4], the Weighted Moving Average and Double Exponential Smoothing method by Hayuningtyas [5] and the Moving Average method by Nurlifa and Kusumadewi [6]. In this study, the focus of the method that will be applied by the author is the Weighted Moving Average with various advantages [7], [8], [9], [10]. Therefore, this study aims to apply the forecasting application model for procurement stock products properly and accurately by applying the Weighted Moving Average method at Dhanty Stores.

II. RESEARCH METHODOLOGY
The method in this study applies the CRISP-DM standard to analyze, design and evaluate the process as shown in Figure 1

A. Identification of Problems
The first step in this research is to identify the object of research and define the existing problems. At this stage, the authors conducted initial interviews with the person in charge of the research object. The results of the identification of these problems are summarized and written as in the background of this research which is then analyzed according to the CRISP-DM standard [11]. This is because the topic and problem in this research is data mining [12].

B. Problem Analysis
Based on the previous problems, the authors conducted a problem analysis by applying the standards and stages of CRISP-DM. The processes and stages are business understanding, data understanding, data preparation, modeling, and evaluation [13].

Business Understanding
The process of understanding the business on the object of research to find out in detail the existing problems. From this process, the author knows the business and solutions to existing problems. The results of the business understanding process become the basis for conducting data understanding as the next stage in CRISP-DM.

Data Understanding
The next step is understanding data. At this stage, the author also collects data related to the research topic. To get the data, the author uses several techniques a. Interview The author again conducted an in-depth interview with Dhanty Store, including a discussion of the details of the existing data such as the origin and types of existing data. b. Observation The authors also made observations to find out an overview of the operational process at Dhanty Store. c. Internal data The internal data in this study is data on product sales in the Dhanty Store. The data will be processed further in solving the problems in this research.

C. Data Preparation
The next step is the preparation of the existing data. This stage is a continuation of the process and standard in CRISP-DM. This stage includes data selection, data cleaning, data reduction and data transformation

Data Selection
At this stage, the authors select and determine the data set that will be used in this study. The data will be processed according to the objectives and needs in this study to forecast the stock of procurement products at the Dhanty Store. The data obtained in this stage is the population in general research.

Data Cleaning
Data cleaning to ensure the data set meets the needs. The data is ensured that it is in good condition and there are no more missing values in it. Missing values include incomplete data, outliers (abnormal data) or data whose values are inconsistent. From the data cleaning stage, data is generated that no longer has a missing value, the data conditions are normal for the next stage.

Data Reduction
Data reduction in data mining is the stage of selecting a sample in the population. The data generated at this stage is real data to be processed using the moving average method in this study.

Data Transformation
At this stage, all data attributes are equated in both format and type. After performing this data transformation stage, the data is ready to be processed.

D. Modelling
The next stage according to the CRISP-DM standard is modeling. At this stage, the author begins to implement the data mining process using the weighted moving average method into a data set that has been prepared previously. The main purpose of this process is to forecast the stock of procurement products at the Dhanty Store. To maximize forecasting results, the author will apply 2 moving averages, 4-week moving average and 5-week moving average [14], [15]. The results of the implementation of the method will be evaluated and tested to determine the best forecasting results.

E. Evaluation
According to the CRISP-DM standard, the stages after the modeling process are evaluation and testing. To carry out the evaluation and testing of the methods and forecasting results of this study, the authors used the Mean Absolute Deviation (MAD) and Mean Absolute Percentage Error (MAPE) methods [16]. The results of this evaluation will determine the best method that can be used in the forecasting process.

F. System Implementation
In designing and developing a prototype application for stock forecasting for procurement of the weighted moving average method at Dhanty Store, the author uses the prototyping model developed by Pressman [17]. In this model, there are five main stages, namely communication, quick plan and quick plan modeling, construction of prototype and deployment delivery & feedback.

G. System Testing
The author tested the system using the System Usability Scale (SUS) method to determine the functionality and usability as well as the level of user acceptance of the resulting application. The researcher gave the SUS questionnaire containing 10 questions to 5 employees at Dhanty Store.

A. Business Understanding
As previously explained, some of the problems that Dhanty Store is currently facing regarding the operational process and product procurement are: 1. The demand and supply of products at the Dhanty Store has not been well controlled so that sometimes the stock runs out while customer demand is high. This causes Dhanty Store customers to move to other stores that offer similar products. 2. The request process until the delivery of products from suppliers takes 3-5 days so that when the stock runs out it will disrupt operations and transactions at the Dhanty Store. This causes a decrease in customer loyalty and even worsens the image of Dhanty Store. 3. The number of requests and product procurement at the Dhanty Store is still as desired without seeing the existing sales data. This causes sometimes goods will pile up in the warehouse or run out altogether.

B. Data Understanding
Based on the results of data collection by interview, observation and document study methods, the authors get raw data from Dhanty Stores to be processed in data preparation. The data is 3000 rows with 15 attributes. The research data samples are as in Table 1.

C. Data Preparation 1. Data Selection
To perform data selection, the author selects the data as needed and adjusts it to the weighted moving average method that will be applied in this study. The author also coordinates directly with Toko Dhanty regarding the selection of this data. As a result, it is determined that the data to be used is transaction data in the last 20 weeks, from the 1st week of August 2021 to the 4th week of December 2021 as many as 3085 rows of data.

Data Cleaning
The author performs a data cleaning process to ensure that the data is in good condition and there are no more missing values in it. From the data cleaning results, it was found that there were 53 incomplete data lines and 1431 inconsistent data lines. The author then cleans the data and equates the format to get the correct data as needed.

Data Reduction
Based on the data cleaning process in the previous stage, finally the existing data was reduced. The data reduction process is done by removing 53 lines of incomplete data so that the total data to be applied to the moving

Data Transformation
In the data reduction process, the existing data is still in the daily time series, while the data requirements for the application of the weighted moving average method are weekly. Therefore, at this stage, the author transforms the data into a weekly time series

D. Attribute and Dataset Selection
Not all data transformation results can be used in the application of the weighted moving average method and algorithm. Therefore, the next step is to select attributes based on the transformation data by looking at existing needs and conditions. As a result, the authors determine two attributes that will be used and applied further in this study, namely the period and the total package.
Thus, the dataset that the author defines is based on the existing data as shown in Table 2.

E. Weighted Moving Average (WMA) Method
The application of the Weighted Moving Average (WMA) method at this stage is a modeling process in CRISP-DM. To get maximum forecasting results, good and accurate conditions with this method, the authors apply 2 WMA algorithms, namely 4 and 5 week moving averages. The aim is to compare forecasting results and select the best algorithm.
Based on the existing dataset, the forecast value (F) of the weighted moving average method is calculated based on the equation: The results of the calculation for forecasting the stock of procurement products at the Dhanty Store with the 4 and 5 week moving average algorithm are as shown in Table 3. error. Later, the quality of the MAD results will be determined by measuring the existing tracking signal, the value is in the range of 4 to -4. As for testing with the MAPE method, there are also several stages, namely calculating the error value and absolute error. The results of this test will determine the best weighted moving average method algorithm that can be used by Dhanty Stores to forecast their product stock in the following weeks. Details of the calculation results of the MAD tracking signal value in each period (t) based on a dataset with a 4-week moving average algorithm as shown in Table 4. Meanwhile, the detailed MAPE value calculation results are based on a dataset with a 4-week moving average algorithm as shown in Table 5. For evaluation results using the MAD and MAPE methods are based on a dataset with a 5-week moving average algorithm as shown in Table  6 and Table 7.

Period Index (t) Actual Data (A) Forecast (F) Error (E) Absolute Error
Absolute Error -Actual Data  4th week of November  16  411  330  81  13  0,1  1st week of December  17  343  378  -35  44  0,4  2nd week of December  18  354  370  -16  150  0,5  3rd week of December  19  341  362  -21  67  0,3  4th week of December  20  170  352  -182  24  0,1  Total  5,7  n  16  MAPE 35,4  Based on the test results on the forecasting value using the weighted moving average method on both algorithms, the 4-week moving average and the 5-week moving average, the data is generated as shown in Table 8. From Table 8, it is known that the 4-week moving average algorithm produces better values based on testing using the Mean Absolute Deviation (MAD) method, which is in the signal range -1.51 to 3.86. This value indicates that the algorithm can be applied properly and accurately for forecasting the stock of procurement products at the Dhanty Store because it is still within the normal point of the MAD signal, which is between -4.00 to 4.00. The tracking signal graph is as shown in Figure 2.   The same result is also seen in the test using the Mean Absolute Percentage Error (MAPE) method where the value for the 4-week moving average algorithm is 35.4% better than the 5-week moving average algorithm of 37.8%. However, in terms of data, actually the difference between the two is not too far, it is still in the same range, which is good and feasible to use.
However, in general, based on testing the MAD and MAPE methods, the 4-week moving average algorithm is better and more accurate to use than the 5-week moving average. So, for the prototype development stage, the author will apply the weighted moving average method with a 4-week moving average algorithm. There are several things that affect the results, including the number of datasets used and the high volatility of the existing data. If the training data is added, the forecasting results may be better in both algorithms. Regarding data fluctuations, this is reasonable because product sales data at Dhanty Stores for clothing and accessories are highly volatile based on trends and customer demand in a period (t).

G. Prototype Implementation
At this stage, the author uses the Unified Modeling Language (UML) tool in general system design. The results of this design can be used as documentation in the development of the application in the future. The use case diagram for the application of stock forecasting for procurement products using the weighted moving average method at the Dhanty Store in this study is as shown in Figure 4.   Figure. 3 Use case diagram The activity diagram of the procurement inventory forecasting application in this study is as shown in Figure 4.

H. Prototype Testing
Testing the prototype using the System Usability Scale (SUS) method which aims to check the reliability and level of user acceptance with the application for forecasting the stock of procurement products using the weighted moving average method. This test uses the SUS questionnaire which contains 10 questions. In addition, testing indicate that the algorithm can be applied well to the forecasting model because it is still in accordance with the signal point threshold, which is between -4 to 4. In addition, the test using the Mean Absolute Percentage Error (MAPE) method shows a value of 35.4 which indicates that the forecasting model algorithm is good and feasible to use.

The development of Forecasting Application with Weighted Moving Average Method for Product
Procurement Stock can make good and accurate predictions on the Dhanty Store according to the needs and can be accepted by its users. Based on the results of testing the prototype using the System Usability Scale (SUS) method, it is known that the average value of the respondents is 82 with details 0% of respondents assessing it is not appropriate, 40% of respondents assessing neutral and 60% of respondents assessing the application is appropriate and running well.