Model Fit Based On Genetic Algorithm And Its Medical Applications

We often expect to show the relations of variables X and Y using a mathematic model which have been obtained from the experiments or observations. So we can use this model to forecast and analysis the one variable from the other. This process to fit linear or nonlinear equation according to measured data is named as curve fitting. The traditional methods for model function are linear least squares, nonlinear least squares and maximum likelihood estimate. They obtain results by numerical iteration method and require that functions must be continuous and differential. Therefore they have been restricted in application. Sometimes the results also get into local optimum. The paper presents a new modelfitting method-genetic algorithms(GA). As a random searchingmethod imitating natural evolutional processes, genetic algorithms starts to search the optimal solution from the population of points. It only needs fitness function information and not differential coefficient and other assistant information. So it ensures that we can get the globally optimal solution.As a good solidity and adaptability algorithm, genetic algorithms has been widely applied in computer science, process control, economic forecast and engineering optimization. But the application in medical is much less. So it is significant to introducing genetic algorithms in medical research.In my paper, the various methods of model fitting are introduced ,especially the principle and steps of genetic algorithms are expounded. Moreover, the indexes used to evaluate model's fitness are also introduced, including coefficient of determination, residual standard deviation and AIC. The paper adopts traditional methods and genetic algorithms to fit five examples, and compares their fitting results. The results show genetic algorithms has good adaptability in linear and nonlinear or continuous variables and discrete variables questions, and besides fitting effect is better than traditional methods.