OPTIMIZATION OF REGIONAL BUDGET ALLOCATION USING GENETIC ALGORITHM AND BAYESIAN HYPERPARAMETER OPTIMIZATION

Abstract

The allocation of budget is a complicated issue that necessitates considering the interests of enterprises, local governments, and citizens. In order to enhance the budget allocation process, this study compares the performance of a hyperparameter-optimized model utilizing Bayesian optimization with a fixed-parameter baseline genetic algorithm (GA) model. Comparative study of the models and the creation of a utility function were among the goals. The findings shown that the improved GA model converges more quickly (30–40 generations versus 40–50) and achieves a higher maximum fitness (5200 vs. 5000 for the baseline), although it takes longer to tune (56 seconds versus 5 seconds).

The approach is promising for managing the allocation process because hyperparameter optimization enhances the quality of budget allocation.  The optimized model achieves greater maximum fitness (7.5-8 units vs. 6.5-7 for the baseline) and converges more quickly (30-40 generations vs. 40-50), according to the convergence plot (Baseline vs. Optimized GA).  A higher-quality and more homogeneous population is shown by the optimized model's population's higher average fitness (6.5–7 versus 5.5–6) and smaller difference between maximum and average fitness