Time Series Forecasting and Modeling of Food Demand Supply Chain Based on Regressors Analysis

Abstract

Demand forecasting has become musty, especially in an industry like food. Short shelf lives per item make badly managed stocks fast detractors for a company. Some machine learning and deep learning techniques showed improvement in recent times at handling a time-series without much improvement of record levels. The study goes into the 'Food Demand Forecasting' dataset released by Genpact and analyzes the effects of different parameters on demand for feature derivation that probably do influence and provides analysis among seven types of regressors for forecasting the order number. This included Random Forest Regressor, Gradient Boosting Regressor (GBR), Light Gradient Boosting Machine Regressor (LightGBM), Extreme Gradient Boosting Regressor (XGBoost), and Cat Boost Regressor. Such a study can demonstrate the efficacy of deep learning models in the forecasting and compare LSTM to other algorithms. The calculated RMSLE, RMSE, MAPE, and MAE values stand at 0.28, 18.83, 6.56%, and 14.18, respectively. These indices would vouch for the prove of the model both against relative and absolute errors. More feature engineering and decomposition of the time series into its additive components are also suggested for performance improvement. In this manner, the proposed system can help the supply chain managers model their resources optimally when making the inventory decisions that will assist in reducing waste.