Vegetable Price Forecasting
This project presents an interactive dashboard for a comparative analysis of time series models (ARIMA, LSTM, SARIMA) used to predict agricultural commodity prices. The goal is to identify the most accurate forecasting approach by evaluating performance on real-world data for tomatoes and potatoes.
View Code on GitHubWinning Model
Cross-Validation Results
Average RMSE across 5 folds. Lower is better.
Model Performance (RMSE)
Root Mean Squared Error on the hold-out test set. Lower values indicate better accuracy.
Forecast vs. Actual Price
ARIMA
LSTM
SARIMA
Conclusion
The analysis clearly demonstrates that for agricultural price datasets with strong seasonal patterns, the **SARIMA model is the most effective**. By explicitly modeling the annual seasonality, it consistently outperformed both the non-seasonal ARIMA and the more complex LSTM network.
This project highlights a key principle in data science: the importance of choosing a model that aligns with the inherent structure of the data. For forecasting tasks in supply chain and logistics, properly specified statistical models like SARIMA can be more powerful and reliable than more complex, generic approaches like neural networks, especially when clear cyclical patterns are present.