Chronic Kidney Disease (CKD) remains a global health challenge, often progressing silently until irreversible damage occurs. Conventional diagnostic tools, such as serum creatinine and estimated glomerular filtration rate (eGFR), lack sensitivity for early-stage detection, delaying timely interventions. Recent advances in machine learning (ML) offer opportunities to leverage electronic health records (EHRs) and complex clinical data for early risk prediction. However, translation into clinical practice faces barriers, including limited data availability, class imbalance, heterogeneity, lack of generalizability, interpretability concerns, ethical risks, and integration challenges. This article critically reviews these shortcomings, focusing on data quality, model development and transparency, and regulatory adherence. We recommend research directions including achieving better interoperability between different sources of data, considering temporal dynamics, accounting for explainable AI approaches, and encouraging cross-disciplines collaboration. Pooling these views, the manuscript presents a roadmap to guide the future toward transparent, ethically appropriate, and clinically accurate AI-assisted approaches to early CKD prediction.