American
International
University-
Bangladesh

Students & Supervisors

Abstract

Agriculture in Bangladesh is heavily affected by climate change, unpredictable rainfall and frequent floods. Accurately predict crop yields become more difficult because of these challenges. Although remote sensing and Artificial Intelligence (AI) have become powerful tools for improving yield prediction worldwide, Bangladesh still faces issues with limited field data and unstable model performance. The dataset used in this study includes yearly records from 1995 to 2024, covering AI model accuracy, prediction error, and the amount of training data available. By analyzing these values, we can understand how AI models have improved or struggled over time, and how the size of the training data impacts prediction quality. This background helps us determine whether AI can offer dependable and accurate yield prediction for the agricultural sector in Bangladesh. A quantitative methodology was applied, where the dataset was pre-processed, normalized, and statistically analyzed to identify performance trends. Descriptive analytics and correlation tests were performed to test relationships between training data size and model accuracy. A deep learning workflow was then designed using CNN–LSTM architectures trained on synthetic remote sensing features aligned with the sequential dataset. Experimental findings show that AI model accuracy differ significantly between years (min: 28.09%, max: 99.43%), indicating vulnerability in traditional approaches. Prediction errors remained high for low-data years, demonstrating a strong negative correlation between training data size and error rates. From previous studies It can be said that models trained with larger datasets (≥75,000 samples) consistently showed higher accuracy and lower uncertainty. Overall, the results highlight the importance of large-scale, high-quality remote sensing datasets in stabilizing model performance and reducing uncertainty in yield prediction. The suggested deep learning approach manifests improved accuracy and hardiness, offering a scalable solution for real- time agricultural monitoring in Bangladesh. Remote sensing and deep learning can support farmers, but consistent data collection and larger datasets are necessary. This research contributes to building climate-resilient agriculture through AI-enabled remote-sensing yield prediction.

Keywords

Publication Details

• Type of Publication:
• Conference Name: 2nd International Conference on Frontiers in Science: Innovation & Technology for Greener Industry (2nd ICFS:ITGI)
• Date of Conference: 15/01/2026 - 15/01/2026
• Venue: Bangladesh University of Engineering and Technology (BUET)
• Organizer: Faculty of Science, BUET