Machine learning and data-driven methodologies are transforming crop recommendation systems (CRS) to enhance agricultural productivity and sustainability. These systems integrate advanced technologies, such as the Internet of Things (IoT), to provide customized crop selection and management recommendations that address challenges from population growth to climate change. Despite considerable progress, the literature highlights key limitations, particularly the reliance on region-specific data and issues related to data quality and availability. This review synthesizes current findings, identifies research gaps, and proposes future directions to improve CRS adaptability and robustness. The outcomes aim to advance ongoing research initiatives and support the development of more efective crop recommendation systems for diverse agricultural environments.
making [13, 14, 15]. These contributions not only underline the potential of advanced AI methodologies but The advent of Agriculture 4.0 marks a transformative also suggest opportunities for their application in agriera in the agricultural sector, characterized by the in- culture, including crop recommendation systems[16, 17]. tegration of advanced technologies, data analytics, and To systematically explore the current state of CRS, this machine learning into traditional farming practices [1]. study employs a structured, four-step literature review As global population growth and climate change present methodology inspired by established frameworks. The unprecedented challenges to food security, there is an ifrst step involves gathering relevant literature from Scourgent need for innovative solutions to enhance agricul- pus, the largest repository of peer-reviewed scientific tural productivity and sustainability [2, 3]. Crop recom- publications, for the period from 2020 to 2024. The search mendation systems (CRS) have emerged as pivotal tools, was guided by predefined keywords and filtering criteleveraging data-driven methodologies to provide tailored ria to ensure the quality and relevance of the selected recommendations for crop selection and management studies. The following keyword combinations were used: based on various environmental and agronomic factors. "crop recommendation system," "crop selection system,"
Recent advancements in deep learning and machine and "machine learning in crop recommendation." Initial learning techniques have significantly contributed to the search results were refined to maintain consistency and progress of CRS research. These advancements are fur- focus by applying language filters (English only), excludther complemented by developments in related fields, ing dissertations, and limiting the timeframe to studies such as computer vision for automated agricultural mon- published within the specified period. A total of 310 artiitoring, robot control for precision farming[4], and EEG- cles were retrieved, forming the basis for further analysis. based classification[ 5, 6] techniques for understanding The collected literature was analyzed descriptively to human-environment interaction [7, 8, 9, 10]. For instance, identify trends and patterns in the field, covering aspects previous work on employing machine learning for com- such as the number of publications per year, types of pubputer vision tasks has demonstrated robust solutions lications, geographical distribution, citation analysis, and for automating complex systems [11, 12]. Similarly, ef- keyword trends. Subsequently, a subset of high-impact forts in integrating deep learning with robot control have papers was selected from the 310 reviewed articles for highlighted novel approaches to autonomous decision- deeper analysis. This selection was based on criteria such as citation frequency, contributions from diverse SYSTEM 2025: 11th Sapienza Yearly Symposium of Technology, Engi- countries, and alignment with key themes like predictive neering and Mathematics. Rome, June 4-6, 2025 modeling, optimization techniques, and region-specific t$errmisosha@amuendiv.m-bainsckerra@.dzun(Siv.L-b.iTsekrrrai.sdsza)(;Ms..aMyaadn@ceurn);iv-biskra.dz recommendations. The focused analysis aimed to identify (S. Ayad) research gaps, novel approaches, and well-established 0009-0002-6586-7724 (M. Mancer); 0000-0002-0017-4962 methodologies. (S. L. Terrissa); 0000-0002-6876-7254 (S. Ayad) The final step involves identifying key limitations in © 2025 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0). the current literature and proposing directions for future research. This section highlights the gaps and challenges faced by researchers in crop recommendation systems and outlines potential areas for advancement. By synthesizing key findings and identifying research gaps, this review seeks to support researchers and practitioners in enhancing the efectiveness and adoption of crop recommendation systems across diverse agricultural environments. articles ofer in-depth, peer-reviewed research. There are also 13 book chapters providing specialized knowledge and 5 reviews summarizing trends and future directions.
This distribution shows the field’s dynamic nature and the role of conferences in quickly sharing findings.
on crop recommendation systems, as shown in Figure 3.
India leads with 249 publications, demonstrating a strong 2. Descriptive Analysis focus on agricultural technologies to address diverse climatic challenges. The United States (14) and Bangladesh This section provides a descriptive analysis of the re- (11) are also significant contributors, followed by Moviewed literature, examining key aspects such as the rocco (6), China (5), Egypt (5), and Sri Lanka (5), reflectyearly distribution of publications, types of publications, ing regional eforts to improve agricultural productivgeographical trends, frequently used keywords, and the ity. Other countries like Algeria, Iraq, and Italy, with influence of notable studies through citation analysis. fewer publications, indicate emerging interest. The involvement of nations from various continents, such as 2.1. Number of Publications per Year Australia, France, and Ethiopia, underscores the worldwide importance of this research, despite diferences in The distribution of publications from 2020 to 2024 re- research capacity and funding. veals a clear upward trend in research interest in crop recommendation systems. As shown in Figure 1, the field 2.4. Citation Analysis began with 22 publications in 2020 and grew steadily to 52 publications by 2022. A notable surge occurred in 2023, with 114 publications, likely due to advancements in machine learning, IoT, and smart agriculture. In 2024, the count slightly declined to 92 but remains significantly higher than in previous years, indicating sustained strong interest.
their citation counts over time. Figure 4 shows steady growth in both publications and citations from 2020 to 2024. The number of documents increased consistently, with a significant rise in 2023, surpassing 90 publications.
Citations also grew sharply starting in 2022, and by 2024, they are expected to exceed 700, reflecting the growing 2.2. Types of Publications impact of AI-driven crop recommendation systems. The surge in citations in 2023 indicates the influence of earThe literature on crop recommendation systems includes lier foundational research, showing the increasing imporconference papers, journal articles, book chapters, and re- tance of AI in agriculture and interdisciplinary interest views, each contributing unique perspectives. As shown in crop recommendation systems. in Figure 2, conference papers lead with 217 publications, reflecting a focus on recent innovations, while 75 journal
Figure 5, illustrating key themes and trends in crop recommendation system research. Prominent keywords such as "crops," "crop recommendation," "crop selection," and "learning systems" reflect the core focus on using AI to optimize crop decisions. Terms like "Internet of Things (IoT)," "precision farming," and "machine learning" demonstrate the integration of data-driven technologies into 3. Related Works agriculture to improve eficiency and sustainability.
Other clusters, including "soil conditions," "fertilizers," This section reviews key studies defining the state-of-theand "agricultural productivity," emphasize the importance art in crop recommendation, illustrating advancements of environmental and resource management in develop- from foundational predictive models to more sophistiing crop recommendation models. Keywords like "de- cated approaches. cision trees," "support vector regression," and "genetic Bhat et al. [18] propose a hybrid model combining Graalgorithms" reveal the variety of machine learning tech- dient Boosted Regression Trees (GBRT) and deep learnniques applied in this field. ing, optimized via Bayesian techniques, achieving an Interdisciplinary themes, such as "economics," "logis- F1-score of 1.0 by leveraging 1,148 soil data points on tics," and "food supply," show the broader socioeconomic dimensions of the research. Additionally, keywords like "climate conditions," "weather prediction," and "soil moisture" highlight the focus on external factors influencing agricultural decision-making. This analysis reflects the merging of AI, environmental science, and agricultural economics in advancing crop recommendation systems.
Addressing uncertainty, [39] applies a Bayesian Belief varieties, making it dificult to apply them across diverse Network (BBN) to adjust crop selection based on local geographic regions. Additionally, data availability and conditions, while [40] proposes a cluster-based system quality remain barriers; most models depend on detailed for crop grouping, optimizing machine learning perfor- parameters like soil nutrients, climate data, and historical mance. crop data, which are often inconsistent or unavailable
In Rwanda, Musanase et al. [30] achieve a 99% train- in certain rural or developing regions. This lack of reing accuracy with neural network models. Kiruthika et liable, high-quality data can significantly afect model al. [29] present a Weighted Long Short-Term Memory performance. (WLSTM) model, reaching 98.35% accuracy by combining Another challenge is the dependency on advanced nutrient data with seasonal patterns. technologies, including IoT for real-time data collection,
Rani et al. [28] explore various models with weather which, while enhancing model precision, requires sigand soil datasets over six years, achieving robust accuracy. nificant investment in infrastructure. This reliance may Janrao et al. [35] similarly use regression techniques to limit the usability of models in areas without such reforecast crop yield and prices, achieving high regression sources. Furthermore, complex methodologies, such as accuracy. deep learning and ensemble techniques, demand high
Numerous studies, including those by [41] and [42], computational power, which can be infeasible for small highlight IoT’s role in real-time data collection, achieving farms or regions with limited technology. Compounding accuracies of 99.14% and 99.55%, benefiting smallholder this issue is the lack of interpretability in many advanced farmers. Abdullahi et al. [37] and Mancer et al. [34] models, leading to a lack of trust among stakeholders. achieve 99.2% and 99.31% accuracy, respectively, using When the reasoning behind AI-based recommendations localized data and ensemble techniques. is not transparent, farmers and decision-makers may
Ensemble methods are further advanced by [43], hesitate to adopt these systems, as they struggle to unachieving 99% classification accuracy, while [ 33] applies derstand and verify the AI’s predictions. Moth Flame Optimization (MFO) to enhance ensemble Future Research Directions: To overcome these limimodel performance, reaching 99.32% accuracy. tations, future research should prioritize the development
To provide a comprehensive analysis, the following of more generalizable models that can adapt across reworks will be examined in a comparative table (Table 3). gions with minimal adjustments. Techniques like transfer
Table 3 compares crop recommendation studies across learning or domain adaptation could enable these models diverse regions, including India, Morocco, Taiwan, and to be applied successfully in diverse agricultural settings. Rwanda, each tailored to specific local conditions. This Furthermore, enhancing data sources by integrating satelgeographic focus, however, limits broader applicability, lite imagery, remote sensing, and publicly available clias models trained on region-specific data may not per- mate data can reduce the dependency on localized data, form well in other areas. Methodologies range from tradi- making models more adaptable and scalable. tional machine learning models to advanced deep learn- Improving data quality and accessibility will also be ing and optimization techniques, such as Neural Net- vital. In addition, future models should focus on beworks, LSTM, and Bayesian optimization, with recent ing lightweight and cost-efective, performing eficiently studies incorporating IoT sensors for real-time monitor- without extensive computational resources—benefiting ing. Advanced models generally show higher accuracy, smallholder farmers in resource-limited settings. often exceeding 97%, with deep learning and ensemble Another promising direction is the enhancement of methods achieving the best results. While data sources model interpretability. By building transparency into include soil nutrients, weather data, and IoT sensor infor- complex algorithms, researchers can make these models mation, dependence on specific regional data and sensor more trustworthy and comprehensible, fostering greater technology impacts scalability. Findings emphasize the acceptance and adoption. Additionally, designing models potential of these systems but underscore challenges in with expandability in mind will ensure that systems can adapting models for varied agricultural environments. evolve to accommodate new data types, crops, and environmental conditions, extending their relevance over time. Finally, integrating blockchain technology could 4. Limitations and Future address data security and traceability concerns, making Research Directions sure that the data used is tamper-proof and transparent.
Addressing these limitations and advancing research in Limitations: A key limitation is the reliance on region- these areas will result in crop recommendation systems specific data, which restricts the generalizability of mod- that are more robust, scalable, and accessible, efectively els. Many systems are finely tuned to local conditions, meeting the diverse and evolving needs of global agriculsuch as specific soil types, climate variables, and crop ture.
ChatGPT, Grammarly in order to: Grammar and spelling check, Paraphrase and reword. After using this tool/service, the authors reviewed and edited the content as needed and take full responsibility for the publication’s content.
This review examines the advancements in crop recommendation systems driven by machine learning and IoT, demonstrating the potential to improve crop selection based on soil, climate, and environmental factors. Despite progress, challenges remain, such as reliance on regionspecific data and the need for high-quality inputs and IoT infrastructure, limiting scalability across regions. Future research should focus on developing more generalizable and explainable models, leveraging transfer learning and remote sensing data to enhance adaptability. Addressing these issues will make crop recommendation systems more robust, accessible, and supportive of agricultural sustainability globally. tional 8 (2024) 104. Analyzing eeg patterns in young adults exposed to [17] A. Tibermacine, N. Djedi, Neat neural networks to diferent acrophobia levels: a vr study, Frontiers control and simulate virtual creature’s locomotion, in Human Neuroscience 18 (2024). doi:10.3389/ in: 2014 International Conference on Multimedia fnhum.2024.1348154.
Computing and Systems (ICMCS), IEEE, 2014, pp. [28] S. Rani, A. Mishra, A. Kataria, S. Mallik, H. Qin,
Ma9–14. chine learning-based optimal crop selection system
[18] S. Bhat, I. Hussain, N. Huang, Soil suitability classi- in smart agriculture, Scientific Reports 13 (
Conferences, volume 430, 2023, p. 01100. [32] M. Mundada, Others, Optimized farming: Crop
[22] A. Villanueva, A. Dorado, E. Gerarman, J. Quebral, recommendation system using predictive analytics,
M. Venal, D. Valenzuela, M. Rosales, Real-time International Journal Of Intelligent Engineering &
best-fitted crops recommendation system based on Systems 16 (