Development of AI-ML based models for predicting prices of agri-horticultural commodities such as pulses and vegetable (onion, potato, onion)
Abstract : The Department of Consumer Affairs monitors the daily prices of 22 essential food commodities through 550 price reporting centres across the country. The Department also maintains buffer stock of pulses, viz., gram, tur, urad, moon and masur, and onion for strategic market interventions to stabilize the volatility in prices.
Decisions for market interventions such as release of stocks from the buffer are taken on the basis of the price trends and outlook. At present, the analyses of prices are based on the seasonality, historical and emerging trends, market
intelligence inputs, crop sowing and production estimates. ARIMA based economic models have also been used to examine and forecast prices of pulses.
Despite its great potential, agriculture price prediction is challenging due to the complex and dynamic nature of the agricultural market, which is influenced by a wide range of factors, including weather variability, supply and demand dynamics, market interdependencies, data availability, and the complexities of agrarian systems [12-20].
Machine learning algorithms have the potential to revolutionize agricultural price prediction [21, 22] by improving accuracy, real-time prediction, customization, and integration.
In this study, we systematically review the stateof-the-art research in agriculture price prediction. We systematize the themes, common problems, approaches, and current progress. Based on that, we identify and recommend future research directions.
1. Data Quality and Availability
Incomplete Data: Historical price data might be incomplete or missing, making it difficult to train accurate models.
Inconsistent Data: Variability in data sources and reporting standards can lead to inconsistencies.
Granularity: Data may lack the granularity needed for detailed predictions (e.g., daily vs. monthly data).
2. Complexity of Models
Model Complexity: AI and ML models, particularly deep learning models, can be complex and require significant computational resources.
Overfitting: Complex models might overfit the training data and perform poorly on unseen data.
Interpretability: Many advanced models (e.g., deep neural networks) are often considered "black boxes," making it difficult to understand how predictions are made.
The system begins with the Data Collection and Integration module, which aggregates data from diverse sources, including historical market prices, meteorological data, soil conditions, and economic indicators. This data is then processed through the Data Preprocessing module, where it is cleaned, normalized, and transformed to create a robust dataset.
Feature engineering techniques are employed to generate relevant predictors that capture temporal patterns and interactions between variables.
In the Exploratory Data Analysis (EDA) phase, the system visualizes trends, correlations, and distributions, providing a clear understanding of the data's characteristics. This analysis informs the selection of appropriate Modeling Techniques.
The system utilizes a range of models, including traditional time series models like ARIMA and SARIMA, machine learning algorithms such as Random Forest and Gradient Boosting Machines, and advanced deep learning methods like LSTM and CNN. These models are trained and validated using historical data, with hyperparameter tuning to optimize performance.
1. Improved Accuracy and Precision
Enhanced Forecasting: AI and ML models can process vast amounts of historical data to generate more accurate and precise price predictions compared to traditional methods.
Complex Patterns: These models can capture complex patterns and relationships in the data that might be missed by simpler forecasting techniques.
2. Real-Time Predictions
Up-to-Date Information: AI and ML models can provide real-time or near-real-time price forecasts, helping stakeholders make informed decisions quickly.
Dynamic Adjustments: Models can be continuously updated with new data to reflect recent market changes and trends.
|