This is the final project for the University of San Diego’s ADS 506 Applied Time Series Analysis course. The technical workbook can be found at the main GitHub repository.
The purpose of this project aims to describe grocery store sales behavior and forecast these sales across a range of product families using data-driven and model-based approaches. The project’s overarching business goal is to establish, in practical terms, the feasibility of using – and potentially automating – one or more of these approaches in a real-world scenario. Note that while the expected outcome is a generalized approach, the study uses representative secondary data from a public data source.
- Data Preprocessing
- Data Visualization
- Exploratory Data Analysis
- Modeling
- RStudio
- The store sales dataset is provided by Kaggle.
- The following model-based and data-driven models are used:
- Naive Forecast
- Holt-Winter's Exponential Smoothing
- Exponential Smoothing (ANN: Additive error, no trend, no seasonality)
- Exponential Smoothing (AAN: Additive error, additive trend, no seasonality
- Exponential Smoothing (MMN: Multiplicative error, multiplicative trend, no seasonality)
- Exponential Smoothing (MMdN: Multiplicative error, multiplicative damped trend, no seasonality)
- Simple Regression
- First-Order Autoregression (AR(1))
- ARIMA
- Neural Network Autoregression
- Clone this repository (For help, refer to this tutorial)
- Raw data is kept in the GitHub repository and Kaggle.
- Data preprocessing, exploratory data analysis, and models are in the GitHub respository.
- Dave Friesen
- Christine Vu