A transformer-based framework for enterprise sales forecasting.

Area of Science:

• Business Analytics
• Machine Learning
• Deep Learning

Background:

• Sales forecasting is crucial for business operations, influencing inventory, resource allocation, and financial planning.
• Accurate sales predictions are vital for optimizing cash flow, adapting strategies, and strategic planning.

Purpose of the Study:

• To present a computational framework for business sales prediction using transformers.
• To tailor a transformer model for low-dimensional tabular data.

Main Methods:

• Developed a novel computational framework utilizing transformer deep learning architecture.
• Designed the model specifically for low-dimensional tabular datasets.

Main Results:

• The proposed transformer model significantly outperformed conventional machine learning models.
• Achieved reduced Mean Absolute Error (MAE), Mean Square Error (MSE), and Root Mean Square Error (RMSE).
• Attained high R-squared values, nearing 0.95, indicating superior predictive accuracy.

Conclusions:

• The transformer-based model is effective for sales forecasting with low-dimensional tabular data.
• The model's accuracy and stability offer a valuable tool for enhancing business decision-making.
• Applicable to various studies involving low-dimensional tabular data analysis.