Transfer Learning in Data-Scarce Agricultural Yield Forecasting: A Bibliometric and Systematic Literature Review
Khoirudin (a*), Sri Yulianto Joko Prasetyo (b), Sutarto Wijono (b), Evi Maria (b), Untung Rahardja (c)

a) Informatics Engineering, Universitas Semarang
*khoirudin[at]usm.ac.id
b) Doctor of Computer Science, Satya Wacana University Salatiga, Indonesia
c) Faculty of Science and Technology, Universitas Raharja

Abstract

Transfer learning (TL) presents a viable approach to enhance the precision of agricultural yield forecasting in data-scarce settings. This study seeks to analyze the advancements, methodologies, and research deficiencies concerning the utilization of TL in agricultural yield forecasting via a Systematic Literature Review (SLR) and bibliometric analysis of 63 Scopus articles from 2020 to 2025. The study was performed with the PRISMA and PICOC frameworks, aided by Biblioshiny in RStudio. The study findings indicate a rising trend in publications beyond 2021, with the predominant transfer learning methodologies being fine-tuning, feature extraction, and domain adaptation utilizing pretrained convolutional neural networks. Research mostly employs satellite images (Sentinel-2) and focuses on nations including China, India, and the United States. Deficiencies were identified in spatial validation, multimodal data integration, and the examination of model security dimensions. This paper offers a literature review and strategic recommendations for advancing AI-driven precision agriculture in data-scarce environments.

Keywords: Transfer Learning- Agricultural Yield Prediction- Remote Sensing- Limited Data- Systematic Literature Review- Bibliometrics- Precision Agriculture- Deep Learning

Topic: Technological and Scientific Innovation in Coastal Cities