Building a Lucy hybrid model for grocery sales forecasting based on time series

Nowadays, time series data are applied in many fields, such as economics, medicine, biology, science, society, nature, environment, or typically in weather forecasting. Time series is a tool that includes methodological formulas and models to help us analyze time series data, extract potentially valuable information, capture historical fluctuations, present and support forecasts of the value of the research object in future. There are many models and methods of time series analysis that have been researched and improved these days for trend analysis and forecasts. Techniques related to time series data processing include linear regression with time series with two features unique to time series lags and time steps, the trend for model long-term changes with moving averages and time dummy, seasonality to create indicators, Fourier features to capture periodic change, and time series as features to predict the future from the pass with a lag embedding. In this article, we build a new hybrid model called Lucy Hybrid that provides full steps in the machine learning process including data pre-processing, training model, evaluation model with Mean Square Error (MSE), Root-Mean-Square Error (RMSE) and Mean Absolute Error (MAE) to compare and get the best model quality. The model also provides functions like storage and loading model to support researchers to reuse and save time on training model. In the Lucy hybrid, we also support the trend and forecast function for time series data. We experiment with a large dataset of more than 3,000,000 records from a large Ecuadorian-based grocery retailer, and we used Linear Regression, Elastic Net, Lasso, Ridge and Extra Trees Regressor, Random Forest Regressor, K-Neighbors Regressor, MLP Regressor, XGB Regressor to experiment and create 20 Lucy hybrid sample models and publish a full source code for researchers to use to expand the model.