پیش بینی میزان فروش با بهره برداری از روش یادگیری متا (مورد مطالعه: فروش آلومینیوم در بورس کالای ایران) (مقاله علمی وزارت علوم)
درجه علمی: نشریه علمی (وزارت علوم)
آرشیو
چکیده
با توجه به جایگاه ویژه پیش بینی فروش بازار در تحلیل فرصت های بازاریابی و نقش مهمی که پیش بینی فروش در برنامه ریزی بخش های مختلف یک سازمان دارد، هدف اصلی پژوهش حاضر استفاده از یکی از ابزارهای نوین حوزه یادگیری ماشین (روش یادگیری متا) برای پیش بینی میزان فروش با مطالعه موردی فروش آلومینیوم در بورس کالای ایران است. پژوهش حاضر در دسته پژوهش های کاربردی قرار می گیرد و داده های پژوهش با استفاده از منبع داده های ثانویه گردآوری شده است. در این پژوهش برای تجزیه و تحلیل از اطلاعات موجود در گزارش های هفتگی منتشر شده در پایگاه رسمی سازمان بورس کالای ایران استفاده شده است. مراحل انجام دادن پژوهش بر مبنای اصول پیش بینی و با استفاده از رویکرد روش یادگیری متا صورت گرفته است. در یافته های پژوهش حاضر چگونگی استفاده از روش یادگیری متا برای پیش بینی میزان فروش و تخمین تقاضای آلومینیوم در بورس کالای ایران نشان داده شد. مدل مبتنی بر روش یادگیری متا برمبنای 4 روش پیش بینی پایه ای شبکه عصبی، آریما، رگرسیون و هموارسازی نمایی و بر بستر داده های سری زمانی مربوط به فروش آلومینیوم در بورس کالای ایران (شامل 344 مقطع زمانی بین سال های 1394 تا 1401) ارائه شده است. بررسی دقت نتایج حاصل از به کارگیری روش یادگیری متا برای پیش بینی میزان فروش، برتری این روش را در مقایسه با چهار روش پیش بینی منتخب دیگر نشان داده است. در این پژوهش برای صحت سنجی نتایج به دست آمده سه مرحله اعتبارسنجی صورت گرفت که در نتایج هر سه نمونه اعتبارسنجی، برتری دقت روش یادگیری متا تأیید شده است. در پژوهش حاضر از روش یادگیری متا برای پیش بینی میزان فروش استفاده شده است. توانایی این روش در حل مسئله پیش بینی یکی از قابلیت های استفاده از ابزارهای هوش مصنوعی را در حل مسائل مختلف مدیریتی نشان می دهد. در نتیجه این پژوهش، روش یادگیری متا به عنوان ابزاری توانمند در حوزه پیش بینی میزان فروش به مدیران بازاریابی و پژوهشگران این حوزه معرفی شده است.Sales Forecasting Using the Meta-learning Method (Case study: Aluminum Sales in Iran’s Mercantile Exchange Market)
Background: Market sales forecasting has a special role in the analysis of marketing opportunities. Along the same line, sales forecasting plays an important role in the planning of different departments of an organization
Objective: Accordingly, the main goal of this research is to use one of the new tools in the field of machine learning (the meta-learning method) to predict the number of sales with a case study. Aluminum is sold in the Iran Mercantile Exchange.
Methods: The current research is included in the category of applied research and the research data were collected using secondary data sources. The data available in the weekly reports published on the official website of the Iran Mercantile Exchange Organization has been used for analysis. The steps of conducting the research are based on the principles of prediction and using the meta-learning approach.
Results: The findings of the current research show how to use the meta-learning method to predict the amount of sales and estimate the demand for aluminum in the Iranian Mercantile Exchange. The model is presented using the meta-learning method based on 4 basic prediction methods of the neural network, ARIMA, regression, and exponential smoothing in line with time series data related to aluminum sales in the Iran Mercantile Exchange (including 344 time periods from 2015 to 2022).
Conclusions: Examining the accuracy of the results of using the meta-learning method to predict sales has shown the superiority of this method compared to four other selected prediction methods. In this research, three stages of validation were conducted to verify the results obtained, and in the results of all three validation samples, the superiority of the accuracy of the meta-learning method was confirmed. The capability of this method to solve the prediction problem is one of the capabilities of using artificial intelligence tools in solving different management problems. As a result, the meta-learning method has been introduced to marketing managers and researchers as a powerful tool in the field of sales forecasting.
Introduction
Market sales forecasting plays a significant role in marketing opportunity analysis. Considering the important role of sales forecasting in the planning of different parts of organizations, the main objective of this research is to utilize a novel machine-learning technique (the meta-learning method) to forecast sales amounts with the case study of aluminum sales in Iran’s mercantile market. In addition, the accuracy of this novel method of forecasting is studied in this research.
Methodology
In this applied research, data collection was done using a secondary data source. The data in weekly reports published in the official Iran mercantile market is used for the analysis. The steps of the research are done based on the forecasting principles and using the meta-learning approach which is based on using meta-knowledge. In this study, the meta-knowledge is achieved by implementing basic methods of forecasting and deriving the results of them. The meta-learning method’s main stage is determining the best approach to use the meta-knowledge in order to achieve the best forecasting model. Due to the consideration of different forecasting results in implementing the meta-learning method, the accuracy superiority of the meta-learning method is expected. The accuracy evaluation is done using practical measurements, and eventually, the results are validated.
Findings
This study showed the use of the meta-learning method to forecast sales, and specifically, to forecast the demand for aluminum in Iran's mercantile market. The approach of the meta-knowledge usage is specifically determined based on the results gathered by analyzing aluminum sales in Iran's mercantile market. It was shown that the meta-learning method is a flexible method that can be used for forecasting sales. In addition, the results of implementing the meta-learning method showed the method’s significant superiority over four other practical forecasting methods. In this research, to validate the results, three steps of validation were conducted. The validation steps were done considering different amounts of input data and different forecasting horizons. Though the accuracy of the forecasting method varied in different validation phases, all three validation results have confirmed the meta-learning method’s superiority in comparison with four other practical forecasting methods.
Conclusions
In this research, the meta-learning method is used to forecast sales, and the accuracy of this method is compared with four other practical forecasting methods. The results of this comparison have shown the capability of this method to solve the forecasting problem. In conclusion, this research has shown the capability of artificial intelligence utilities to solve different management problems, and this could lead to finding new superior tools for managers.
تبلیغات
