jawharjoe / sarima_-sales_forecasting

An in-depth time series analysis repository containing a Jupyter Notebook that demonstrates the application of the SARIMA model to forecast weekly sales data.

SARIMA_-Sales_Forecasting

An in-depth time series analysis repository containing a Jupyter Notebook that demonstrates the application of the SARIMA model to forecast weekly sales data.

Data Import and Preprocessing

• Dataset loaded from CSV, containing weekly sales data from 2008 to 2021.
• Data split into training and testing sets, with 7.5% as test size.

Exploratory Data Analysis

• Time series plot generated to show sales trend.
• Augmented Dickey-Fuller test conducted, p-value ~ 1.10e-15, indicating stationarity.

Model Selection

• Fitted range of SARIMA models to training data.
• Best model selected based on Akaike Information Criterion (AIC), lowest AIC: 14016.79.

Model Diagnostics

• Diagnosed using plots of standardized residuals, histogram, normal Q-Q, and correlogram.
• Residuals are normally distributed, suggesting good fit.

Forecasting

• Model used to forecast future sales for 52 steps ahead.
• Forecasted values compared with observed sales, showing close match.

Model Evaluation

• Performance evaluated using MAPE, MAE, and RMSE:
  • MAPE: 0.4593
  • MAE: 24095.55
  • RMSE: 138524.43

Conclusion

• SARIMA model effective in forecasting weekly sales.
• Can be used for future sales predictions, aiding in planning.