Nonlinear prediction of gross industrial output time series by Gradient Boosting

Predicting gross industrial production is helpful to design plan in development zone. History data in Jinchuan district, Hohhot, were collected. BDS, Ljung-Box, Box-Pierce, White's and Teraesvirta's neural network test and surrogate data test were combined to selecting a proper model. According to phase space reconstruction, function fitting was finished by Gradient Boosting. The results showed that nonlinear dependence existed in series. The production in 2015 was predicted to be 6937977 ten thousand Yuan.