Predictive Intelligence in Public Administration A Machine Learning Framework for Decision Support and Policy Optimization

Abstract

Data-driven LGT can help identify meaning and reshape the provision of public services, and facilitate expansion in new global and digital business sectors. Hence, the value of machine learning (ML)-based predictive intelligence has increased in the context of providing actionable predictions based on real-life data and policy recommendations. In this paper, a comprehensive machine learning framework is proposed to support the decision-making and policy-making processes in a more advanced manner. The framework includes several prediction algorithms such as Linear Regression, Decision Trees, Random Forest and Artificial Neural Networks to make predictions about the important information provided by the government. The models are demonstrated and analyzed using real-world data from open government portals across different scenarios, like prediction of budget, hospital demand forecasting, as well as emergency source allocation. They are tested for their predictive efficiency with traditional metrics such as the Root Mean Square Error (RMSE), the Mean Absolute Error (MAE) and the Coefficient of Determination (R²). A comparison of the experimental results indicates that Artificial Neural Networks have significant predictive performance, especially for complex and high-dimensional data, whereas Random Forest has a high predictive performance and a better interpretability. Linear Regression and Decision Trees are more useful in terms of transparency and scalability, making them appropriate models for selection of public policies in government systems. They are not a good environment for implementation everywhere, however, including universities, NGOs, or national/international government institutions with many requirements for their implementation. This encompasses application of explainable artificial intelligence (XAI) methods to tackle transparency and trust concerns, in addition to designing more local and adaptive models to enhance generalizability to various administrative contexts.