Acute Food Insecurity (AFI) remains a significant and persistent challenge, and Machine Learning (ML) offers promising solutions to improve predictions and early warning systems. ML can integrate large and diverse datasets, considering multiple drivers of AFI, to enhance forecasting capabilities. This study carries out a comprehensive review of existing data, models and applications for predicting the current or future levels of food insecurity in specific regions. Forecasting AFI with ML is a relatively new and underutilised field, facing challenges such as complex data interactions, heterogeneous input variables, Food Security conceptual complexity highlighting the need for cross-disciplinary expertise. Despite these challenges, ML has the potential to provide actionable information for humanitarian intervention, but requires careful data review and consideration of statistical and machine learning methods, as well as explainability techniques, to effectively utilise the increasing amount of available data and ensure reliable model outputs. 

Key Findings and Recommendations:  

- Target variables for AFI ML modeling are scattered and have uneven coverage across sources, highlighting the need for standardisation. 

- A set of optimal input feature sources is proposed to develop actionable models, including past food insecurity and agroclimatic-economic drivers. 

- The XGBoost model is identified as a robust and accurate prediction approach, offering a good tradeoff between complexity and performance. 

- Past food insecurity and agroclimatic-economic drivers have a significant influence on model predictions, emphasizing their importance in AFI forecasting. 

Overall Impact:

The review provides a comprehensive assessment of current ML modeling efforts for AFI forecasting, highlighting possibilities and limitations. By detailing the target variables landscape, recommending optimal input variables, outlining the modeling workflow, and advocating for improved usability, the review aims to support the development of more effective ML models for AFI forecasting, ultimately contributing to better decision-making and early warning systems.