Sales forecasting using Sequential Long short-term memory (LSTM) Model
Abstract/Description
In the world of uncertainty, the evolution of artificial intelligence helps the companies towards more sustainable sales/ demand forecasting. Before the rapid adaptation of Artificial intelligence (AI), forecasting was done with the help of statistical methods such as simple moving averages, exponential smoothing, and others. However, after the development of AI Algorithms and the advancement of computing, companies are using AI for forecasting. AI methodology usually uses historical data to forecast sales/demand. Despite having an advanced AI system in place forecasting errors still exist due to several factors, including data fluctuation, rapid change, and other external factors. The forecasting is performed to avoid unnecessary stock buildup or out-of-stock issues. In this paper, we use historical sales data of medium-sized Departmental Stores. Initially, 10 articles are selected using the ABC analysis to cover high-value article sales forecasting. The sequential Long Short-Term Memory Model (LSTM) is used for forecasting daily sales. The results of AI-generated forecasting sales data are compared with Actual sales for the month to check the efficiency of the system. The result is very promising and MAD (Mean Absolute Deviation) ranges from 6 to 3. Although other models are available, LSTM is used due to computational constraints. The actual data contains some noise data due to out-of-stock and bulk orders. Which might be the leading cause of error as well as seasonal variation.
Keywords
Sales Forecasting, Sequential LSTM Model, Time Series Forecasting, Artificial Neural Networks
Track
Management
Session Number/Theme
1B: Management
Session Chair
Dr. Muhammad Ayaz ; Dr. Muhammad Imran
Start Date/Time
30-5-2024 1:50 PM
End Date/Time
30-5-2024 3:20 PM
Location
MCS-4, AMAN-CED, First Floor
Recommended Citation
Ashraf, S., Ijaz, K., Saleem, M. D., & Sajjad, S. (2024). Sales forecasting using Sequential Long short-term memory (LSTM) Model. 3rd IBA SBS International Conference 2024. Retrieved from https://ir.iba.edu.pk/sbsic/2024/program/10
COinS
Sales forecasting using Sequential Long short-term memory (LSTM) Model
MCS-4, AMAN-CED, First Floor
In the world of uncertainty, the evolution of artificial intelligence helps the companies towards more sustainable sales/ demand forecasting. Before the rapid adaptation of Artificial intelligence (AI), forecasting was done with the help of statistical methods such as simple moving averages, exponential smoothing, and others. However, after the development of AI Algorithms and the advancement of computing, companies are using AI for forecasting. AI methodology usually uses historical data to forecast sales/demand. Despite having an advanced AI system in place forecasting errors still exist due to several factors, including data fluctuation, rapid change, and other external factors. The forecasting is performed to avoid unnecessary stock buildup or out-of-stock issues. In this paper, we use historical sales data of medium-sized Departmental Stores. Initially, 10 articles are selected using the ABC analysis to cover high-value article sales forecasting. The sequential Long Short-Term Memory Model (LSTM) is used for forecasting daily sales. The results of AI-generated forecasting sales data are compared with Actual sales for the month to check the efficiency of the system. The result is very promising and MAD (Mean Absolute Deviation) ranges from 6 to 3. Although other models are available, LSTM is used due to computational constraints. The actual data contains some noise data due to out-of-stock and bulk orders. Which might be the leading cause of error as well as seasonal variation.
Comments
In the updated Manuscript, Table 1~3 has been updated. Further, a detailed comparison of ESM SAM and Proposed LSTM is presented. (Figure # 2)