Abstract—The field of agriculture continues to face challenges in terms of predicting crop yields, which is a critical factor for decision-making at international, regional, and local levels. Crop yield prediction depends on factors such as climatic conditions, soil properties, crop-specific traits and environmental influences. Therefore, selecting the most appropriate crop for maximizing crop yield plays a vital role in improving real-life farming scenarios. In this manuscript, a Class Dependent Label Smoothing-Bayesian Long Short-Term Memory (CDLS-BLSTM) approach is developed for an efficient crop recommendation and yield prediction. The CDLS regularization method is incorporated in the traditional BLSTM technique to prevent the bias and enhance both the recommendation and prediction process. The BLSTM learns distribution across LSTM weights instead of fixed weights and predicts the uncertainties in data. In pre-processing phase, the min-max normalization technique is utilized to scale data into a uniform range of 0 to 1. The developed CDLS-BLSTM acquired an accuracy of 99.78% in Crop Recommendation Dataset (CRD, 2000 samples) dataset and an accuracy of 95.32% in Crop Yield Prediction (CYP, 4000 samples) dataset when compared with conventional algorithms.
 
Keywords—Bayesian Long Short-Term Memory (BLSTM), class dependent label smoothing, crop recommendation, crop yield prediction and min-max normalization


Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).