Sales Forecasting Project

Problem Statement

The objective of this project was to forecast monthly champagne sales for Perrin Freres using time series analysis techniques. The sales data was obtained from the Kaggle dataset: Perrin Freres Monthly Champagne Sales.

Steps Performed

Data Preparation and EDA

• Loaded the dataset using Pandas.
• Renamed columns for clarity.
• Converted the 'Month' column to datetime format and set it as the index.
• Removed the last two rows(Had missing values) for consistency.

Stationarity Check

• Conducted the Augmented Dickey-Fuller (ADF) test to test for stationarity.
• Found that the data was non-stationary, indicating the presence of a unit root.

Differencing

• Applied differencing to make the data stationary.
• Checked for stationarity again using the ADF test, confirming stationarity.

Auto Regressive (AR) Model

• Plotted autocorrelation and partial autocorrelation functions to determine AR model parameters.
• Implemented ARIMA model with order (1,1,1) for forecasting.
• OBSERVATION: Data was originally seasonal(non-stationary), even after converting it to stationary, the forecasting is poor as illustrated in the above graph.

Seasonal ARIMA Model

• Utilized a Seasonal ARIMA (SARIMA) model with seasonal order (1,1,1,12) for improved forecasting.

Forecasting

• Predicted sales for the next 2 years using SARIMA model.