Dynamic Density Adjusted K-Boost Algorithm for Public Finance Risk Management and Decision Support

Managing public finance risk is a great challenge due to the varying nature of financial systems. Situations like uneven data distributions, interconnected variations, and unexpected external shocks are some of the examples. Most of the existing research introduced new insights to solve these types of financial risks. However, the issues regarding these types of tasks continue. This study addresses this type of issue by combining the advantages of supervised and unsupervised learning models. K-means clustering is one of the effective techniques most commonly used to handle these types of complexities, as highlighted in previous research. But the struggles regarding uneven cluster sizes and varying densities make these models a failure. The existing supervised learning models also cannot identify hidden patterns without labeled data. By considering these limitations as a heavy background, we present this study. The proposed approach of the study is novel, developed using the traditional background of K-means clustering and supervised learning models. To avoid the traditional limitations, we proposed an advanced Dynamic Density Adjusted K-Boost (DDA-K-Boost) Algorithm, otherwise known as DKB. This framework combines the advantage of density-aware clustering ability and the predictive power of XGBoost. Here, DKB first applies the DDA-K-Means clustering technique to classify public finance data into clusters based on important similarities. This helps to address the challenges of uneven distributions and hidden variability. Then these clusters are used as the input of XGBoost to predict financial risks like revenue losses and debt growth. The simulation result highlights that the proposed approach achieved the improved clustering precision of 94–97% and a predicting efficacy of 97.32%. Whereas, the wealth ratio indicates the investors remain stable with the improved range of 0.02–0.075. By using the value-at-risk approach, the suggested model’s best consumption stability does not exceed 4% of the total investment wealth; this highlights the risk reduction and decision-making support.