Demands and Sales Forecasting for Retailers by Analyzing Google Trends and Historical Data

Demands and Sales Forecasting for Retailers by Analyzing Google Trends and Historical Data

Md Rokon Uddin, Saman Hassanzadeh Amin, Guoqing Zhang
DOI: 10.4018/978-1-7998-3805-0.ch003
(Individual Chapters)
No Current Special Offers


A supply chain includes several elements such as suppliers, manufacturers, retails, and customers. Forecasting the demands and sales is a challenging task in supply chain management (SCM). The main goal of this research is to create forecasting models for retailers by using artificial neural network (ANN) and to enable them to make accurate business decisions by visualizing future data. Two forecasting models are investigated in this research. One is a sales model that predicts future sales, and the second one is a demand model that predicts future demands. To achieve the mentioned goal, CNN-LSTM model is used for both sales and demand predictions. Based on the obtained results, this hybrid model can learn from very long range of historical data and can predict the future efficiently.
Chapter Preview

2. Literature Review

Prediction of future demand and sales has been a very helpful tool for business decision-makers. Recently, ANN has gained significant attention in Supply Chain Management (SCM) because of its capability to predict future, to process large datasets, to handle very complex non-linear functions, and for its efficiency and robustness in prediction. ANN is helpful even if the data of the problem is partially present.

There are some researchers who have applied ANN to design their own prediction models to forecast demands and sales. Aburto and Weber (2007) developed an inventory management system for a supermarket and explained a hybrid intelligent model for forecasting demand in SCM. This hybrid model is a combination of autoregressive integrated moving average and neural network models. Yin et al. (2008) introduced an adaptive neural network model that had more accurate forecasting results than the traditional neural network. Amin-Naseri and Tabar (2008) applied a comparative study of various neural network models, and concluded that Recurrent Neural Network (RNN) model has the most precise forecasting results.

Google Trends and searches have been utilized in some investigations. Su (2008) analyzed the impact of the ease of online searches on consumers' online search intentions, and showed that there is a noticeable positive impact of cross-site and in-site searches on both priced and non-priced item searches. Ginsberg et al. (2009) revealed that Google Trends was able to trace and predict the spread of influenza earlier than the Centers for Disease Control and Prevention. Choi and Varian (2009) claimed that Google Trends data can be utilized for predicting unemployment rate. They showed in their later research (Choi and Varian, 2012) that Google Trends has an important connection to car or house sales. Shimshoni et al. (2009) explained predictability of Google Trends data itself.

Goel et al. (2010) illustrated some restrictions of search data. They mentioned that search data is helpful in making predictions, but the predictability may not increase noticeably. Guzman (2011) utilized Google search data to predict inflation. Baker and Fradkin (2011) investigated how job search respond to extensions of unemployment payments using Google search data.

Kandananond (2012) compared two data mining methods including ANN and Support Vector Machine (SVM), and used these models to predict the demand of consumer products. They concluded that the SVM had more accurate results in term of Mean Absolute Percentage Error (MAPE) than the ANN. Jun et al. (2014) analyzed that Google Search provides an outstanding platform for observing consumers' activities of information seeking. It reflects the needs, demands, and interests of the customers. As a result, customer preferences can be predicted.

Complete Chapter List

Search this Book: