[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"item-2509":3},{"id":4,"title":5,"url":6,"summary":7,"summary_zh":8,"content":9,"source_name":10,"source_url":6,"published_at":11,"category":12,"cover_url":9,"hotness":13,"is_selected":14,"score":15,"score_detail":16,"sources":24,"tags":26,"view_count":32,"doi":33,"paper":34,"created_at":49},2509,"Artificial Intelligence-Based Crop Recommendation Using Soil and Climate Data: A Comprehensive Review of Machine Learning, Deep Learning, and Smart Agriculture Approaches","https:\u002F\u002Fdoi.org\u002F10.64388\u002Firev10i3-1722922","Crop recommendation systems that integrate soil and climate data with artificial intelligence (AI) have expanded rapidly since 2020, spanning classical machine learning (ML), ensemble methods, deep learning, explainable AI (XAI), and Internet of Things (IoT)-enabled sensing. This review critically synthesizes 35 sources — 33 peer-reviewed journal articles and conference papers plus 2 preprints retained only for background context — comprising 12 studies verified in full against their primary text and 23 studies verified at the bibliographic level, to examine what has been attempted, which data and algorithms have been used, and how reliable the reported results are. The reviewed literature shows convergent use of a narrow feature set (nitrogen, phosphorus, potassium, temperature, humidity, pH, and rainfall) and recurring near-ceiling accuracy, including cases at or above 98% [1], [5], [6] and, in one case, a reported 1.00 across accuracy, precision, recall, and F1-score following class-balancing [7]. Cross-examination of dataset descriptions across studies reveals inconsistent provenance and documentation: structurally similar seven-feature datasets are described with different national contexts and reported sample sizes ranging from 2,100 to 3,000 records [5], [6], [9]. Validation practice is dominated by random hold-out or k-fold cross-validation; among the studies examined in full, only one tested spatial cross-validation, reporting a substantial performance decline (AUC 0.89 to 0.55–0.62) relative to random-split results [4]. Explainable AI and uncertainty quantification remain minority practices in the reviewed literature. A prior conference proceedings paper self-described as a","自2020年以来，将土壤和气候数据与人工智能（AI）相结合的作物推荐系统迅速扩展，涵盖经典机器学习（ML）、集成方法、深度学习、可解释人工智能（XAI）以及物联网（IoT）赋能的传感技术。本综述批判性地综合了35个来源——33篇同行评审期刊论文和会议论文，以及2篇仅用于背景参考的预印本——其中包括12项经全文核实的研究和23项经书目层面核实的研究，以考察已尝试的研究方向、所使用的数据和算法，以及所报告结果的可靠性。所综述的文献显示，特征集使用趋同且范围狭窄（氮、磷、钾、温度、湿度、pH和降雨量），准确率反复接近上限，包括达到或超过98%的案例[1], [5], [6]，以及一例在类别平衡后准确率、精确率、召回率和F1分数均报告为1.00的研究[7]。对各项研究中数据集描述的交叉审查揭示了来源和文档记录的不一致：结构相似的七特征数据集被描述为不同的国家背景，报告的样本量从2,100到3,000条记录不等[5], [6], [9]。验证实践以随机留出法或k折交叉验证为主；在经全文审查的研究中，仅有一项测试了空间交叉验证，报告称相对于随机划分结果，性能显著下降（AUC从0.89降至0.55–0.62）[4]。可解释人工智能和不确定性量化在所综述文献中仍属少数实践。一篇先前会议论文集中自述为",null,"Iconic Research and Engineering Journals","2026-09-14T00:00:00Z","论文",10,false,61,{"impact":17,"substance":18,"depth":19,"authority":20,"freshness":21,"relevant":22,"comment":23},12,18,16,6,9,1,"系统综述揭示作物推荐模型普遍存在数据集来源混乱与验证方法单一（仅一项空间交叉验证即大幅掉点）的问题，对智慧农业AI落地有实质警示价值，但期刊影响力有限。",[25],{"name":10,"url":6},[27,28,29,30,31],"智慧农业","农业人工智能","精准农业","作物推荐","土壤数据",0,"10.64388\u002Firev10i3-1722922",{"doi":33,"openalex_id":35,"authors":36,"venue":10,"cited_by_count":32,"oa_url":9,"card":41,"direction":47,"ingested_from":48},"W7212594050",[37,39],{"name":38,"orcid":9},"Snehal Sanjay Raut",{"name":40,"orcid":9},"Yogesh V. Chimate",{"tldr":42,"method":43,"finding":44,"direction":45,"opportunity":46},"综述AI作物推荐研究，指出高准确率多源于数据与验证缺陷。","系统综述35篇文献，对比ML、DL、XAI与IoT方法及验证方式。","常用7特征数据集来源不一，随机验证致准确率虚高，空间验证性能骤降。","农业人工智能与决策模型","需建立标准化数据集并推广空间交叉验证与不确定性量化，提升模型真实泛化能力。","智慧农业 \u002F 农业物联网","openalex","2026-09-15T23:30:08.397298Z"]