Hybrid Machine Learning Models for E-Commerce Demand Forecasting Using Instacart Dataset

Demand forecasting is a key component of several business operations, enabling businesses to reduce waste, meet the customer’s needs, and optimize their inventory. Given the rapid growth of the e-commerce industry, demand forecasting has evolved into one of the most important requirements for an e-commerce company. Traditional forecasting methods usually fail to capture the dynamic nature of consumer behavior. Similarly, individual machine learning models, while powerful, face limitations such as overfitting, sensitivity to high-cardinality categorical variables, or heavy computation. To address these challenges, this paper proposes a novel hybrid model by fine-tuning and stacking the strengths of the LightGBM, CatBoost, and a Multilayer Perceptron models into a single Ridge regression meta-model for e-commerce demand prediction. Upon evaluation, the proposed model was found to have achieved a mean absolute error of 0.697, a mean squared error of 1.67, and an R2 score of 0.656, outperforming all the individual models. The obtained results demonstrate the ability of the proposed model in capturing complex patterns in e-commerce demand data, including non-linear relationships, temporal dependencies, and categorical feature interactions.