Relational analysis model of weather conditions and sales patterns based on nonnegative tensor factorization

Sei Okayama, Haruka Yamashita, Kenta Mikawa, Masayuki Goto, Tomohiro Yoshikai

Research output: Contribution to journalArticle

Abstract

It is necessary to analyze the relationships between the retail sales of various items and weather conditions. However, the relationship between the sales of each item and the weather condition may vary among stores. Additionally, it is necessary to model the statistical relationships between a wide variety of goods and weather conditions by using past sales data. In such a case, it becomes unrealistic to construct a forecast model for every individual item owing to the breadth of items and the number of retail shops. This study proposes a model to analyze the relationships between the sales of various items and weather conditions. This method can be used to decompose the data into three matrices based on the nonnegative tensor factorization (NTF) method. The results of the analysis clarified that the proposed model can identify important items whose demand is strongly influenced by weather conditions, thereby increasing the effectiveness of inventory management. Additionally, the store clusters estimated by the proposed model can facilitate the construction of regression models that demonstrate the relationship between the sales of each item and weather conditions.

Original languageEnglish
Pages (from-to)2477-2489
Number of pages13
JournalInternational Journal of Production Research
Volume58
Issue number8
DOIs
Publication statusPublished - 2020 Apr 17

Keywords

  • business analytics
  • clustering
  • nonnegative matrix factorization
  • nonnegative tensor factorization
  • weather
  • weather data

ASJC Scopus subject areas

  • Strategy and Management
  • Management Science and Operations Research
  • Industrial and Manufacturing Engineering

Fingerprint Dive into the research topics of 'Relational analysis model of weather conditions and sales patterns based on nonnegative tensor factorization'. Together they form a unique fingerprint.

  • Cite this