Deep learning models were developed to predict agricultural land suitability and soil quality for precision agriculture. TabTransformer achieved the highest accuracy (97.40%) on augmented data, demonstrating the potential of deep learning for structured agricultural data.
This study presents a deep neural network and transformer-based framework for classifying Bengali texts into saint and common forms, with SahajBERT achieving the best performance. The integration of LIME enhances interpretability, providing insights into stylistic cues aligned with linguistic expectations.
LIME was used to enhance the interpretability of model predictions, especially in cases of misclassification, contributing to the development of reliable and interpretable AI-driven solutions for medical imaging.
Handwritten character recognition is a significant research area, especially for languages like Bengali where quality OCR applications are scarce. Merging machine learning and deep learning techniques, including CNN, improves accuracy. The use of Grad-CAM as an XAI method enhances model interpretability.
This paper introduces a GAN-BERT based model for Bengali text classification, addressing the challenge of limited labeled data, and demonstrates its superior performance over traditional BERT models on hate speech and fake news datasets, particularly with small training samples.
This work uses Hidden Markov Models (HMM) to predict next-day stock prices based on historical data, focusing on fractional price changes and intraday highs and lows. The Maximum a Posteriori HMM method showed satisfactory results and can generalize to predict stock prices for any company with proper training.
This paper presents a Generative Adversarial Network (GAN)-based technique that generates new Jamdani textile patterns from user-input sketches, mimicking traditional designs. We created a unique Jamdani motif dataset and achieved satisfactory results using the pix2pix model, potentially advancing research in this area.
