Study Of Archaeological Sites Predictive Distribution Based On Logistic Regression Optimization Method

The distribution of ancient sites is closely related to the natural environment and human environment and the changes of these environmental variables also affect the development of human civilization.It is a trend to build archaeological sites predictive model in recent years.The relationship model of the site and various environmental variables is established to predict the occurrence probability of the site in the survey area by means of statistical and GIS spatial analysis tools.The aim is to reduce the potential damage to the site and provide some scientific basis for archaeological research.Logistic regression model is one of widest model used in sites prediction.Due to the second item classification and multivariate analysis.However,Logistic regression is easy to underfit and the classification accuracy may not be high in classification of predictive sites.In this paper,we will use the modified optimization algorithm to Logistic regression and extract the human environment variables and natural environment variables in target areas by GIS spatial analysis method.By means of Numpy and Matplotlib in Python,we utilize the gradient ascent method commonly used for Logistic regression and modified stochastic gradient ascent method to calculate the model's regression coefficient to establish the sites predictive model of Longshan period in Fenhe River Basin.The predictive model's training precision and prediction precision established by the gradient ascent method is respectively 71.4% and 68.8%.The predictive model's training precision and prediction precision established by the modified stochastic gradient ascent method is respectively 89.2% and 85.3%.By force of contrast the predictive results,it shows that the precision of the modified stochastic gradient ascent method is 16.5% more than the gradient ascent method.Besides,the validation of those two predictive model proposed by the Kvamme gain statistical validity verification is conducted,whose G value is respectively 0.67 and 0.75.It indicates that those two models both have a higher practical application significance.Especially,the logistic regression model established by the modified stochastic gradient ascent method is more significant.