Abstract:Biochar is a carbon-rich solid produced from biomass or organic solid waste via thermochemical conversion and has attracted extensive attention across agriculture, forestry, environmental remediation, energy, chemical engineering, and materials science. Owing to its high carbon content and stability, biochar can serve as an effective solid fuel and a promising material for carbon sequestration. Its porous structure and tunable surface functional groups also make it highly effective in applications such as adsorption, catalysis (including catalyst supports), soil amendment, and hard-carbon anodes for energy storage systems. Despite these advantages, traditional biochar development still depends heavily on empirical trial-and-error and labor-intensive experiments. This approach faces major challenges arising from the wide diversity of biomass feedstocks, the complexity of carbonization and activation parameters, and the limited controllability of biochar structure and performance. These factors hinder rapid preparation, precise design, and efficient scale-up. In recent years, machine learning has become a core tool for data-driven modeling, prediction, and optimization, and it has been increasingly applied to biomass thermochemical conversion, biochar preparation, and biochar-based applications. By establishing predictive links among feedstock characteristics, process conditions, structural properties, and application performance, machine learning helps identify key governing factors and clarify their influence patterns. It can rapidly predict structural features such as pore structure, surface area, degree of graphitization, and surface functional groups, as well as application-oriented properties including adsorption capacity, catalytic activity, electrochemical performance, and environmental functionality. More importantly, through forward prediction, inverse design, and multi-objective optimization, machine learning offers powerful strategies for intelligent feedstock screening, process-parameter optimization, and dynamic control, enabling the targeted design of high-performance biochar tailored to specific applications. This review systematically summarizes the main application areas of biochar; the descriptor systems used to characterize feedstocks, processing conditions, and biochar properties; and recent advances in modeling, performance prediction, and intelligent design strategies driven by machine learning. Particular emphasis is placed on how machine learning can reveal hidden relationships among composition, structure, and performance, accelerate the transition from empirical experimentation to rational design, and substantially improve research and development efficiency. We also discuss current limitations in the literature, including insufficient quantity and quality of data, descriptor inconsistency, weak model interpretability, and limited cross-scale generalization from laboratory systems to pilot or industrial processes. Overall, this review aims to provide a comprehensive reference and strategic guidance for accelerating machine learning-driven biochar preparation, optimization, and applications, and to promote the transition of biochar materials from laboratory research to engineering practice. Close-