In the evaluation process of the comprehensive index of marine economic development in Fujian Province, there are often multicollinearity of indicators and nonlinear problems of regression, which may lead to the model evaluation results not being able to objectively reflect the actual economic development law. The traditional composite index with ocean GDP was replaced as an explanatory variable, the random forest algorithm was used to reduce the collinear influence of the index in the modeling process, and four multilayer perceptron network models to verify the nonlinear relationship in the regression task was established. The results show that there is a significant nonlinear relationship between the development level of marine economy in Fujian Province and the subsystem indicators constructed. The regression model constructed by combining the random forest algorithm with the multilayer perceptron can fit the development of the marine economy in Fujian Province, and there are obvious shortcomings in scientific and technological innovation and ecological environment of the marine economy of Fujian Province.