Enterprise Financial Risk Prediction based on Adaptive Genetic Algorithm and MLP

Aiming at the problem of insufficient prediction accuracy of traditional genetic algorithms in current enterprise financial risk prediction models, a hybrid prediction model (AGA-MLP) based on adaptive genetic algorithm (AGA) optimized multi-layer perceptron (MLP) is proposed. By designing a dynamic crossover probability and mutation probability adjustment mechanism, the genetic parameters are adaptively adjusted according to the population fitness variance, and a prediction model with 15 input nodes and 2 hidden layers is constructed in combination with the MLP network. After optimizing the initial weights and thresholds of MLP using AGA, the momentum accelerated BP algorithm is used for iterative training. On a test set containing financial data of 3,000 listed companies, the model achieved an accuracy of 93.7%, a recall of 91.2%, and an F1 value of 92.4%, and reduced the number of convergence iterations to 65, shortening the training time. This method effectively solves the problems of low efficiency and local convergence of traditional model parameter optimization, and provides a more reliable decision-making basis for risk warning of financial institutions.