BI-LSTM MODELS FOR OPTIMIZED CROP SELECTION AND YIELD FORECAST IN PRECISION AGRICULTURE

Abstract

Agriculture is necessary to feed the world's population, and with climate change and limited resources, it is becoming more and more important to pick the correct crops and guess how much they will produce. This research introduces a deep learning framework that employs Bidirectional Long Short-Term Memory (Bi-LSTM) networks to enhance decision-making in crop selection and accurately predict crop yields. The model incorporates historical data on farming, such how crops have done in the past, the weather (temperature, rainfall, and humidity), and the soil's characteristics. Bi-LSTM is better than regular machine learning models and unidirectional LSTMs since it can find temporal relationships in both directions. The proposed system is trained and evaluated using publicly accessible agricultural information. It can precisely predict how much of a crop will grow and advise the best kind of crops for various places. The study used the Agricultural Crop Yield dataset to train and test the Bi-LSTM model. To discover the Mean Absolute Error, Root Mean Squared Error, R² score, and Mean Absolute Percentage Error, the performance is compared to various methods like Linear Regression, Random Forest, and a basic LSTM. The Bi-LSTM model works best, with an MAE of 0.32, an RMSE of 0.47, and a R² score of 0.91. This study's findings may assist farmers, agronomists, and policymakers in optimizing resource use and enhancing agricultural productivity using data-driven insights.