| Creativity is the source of national scientific and technological progress and economic development.In the present,individual creativity is generally measured and evaluated through various questionnaires.This measurement and evaluation process may be mixed with a lot of subjectivity and uncertainty.In recent years,scholars have begun to study the possibility of predicting individual creativity through the use of functional connections in the brain.However,the prediction model uses a large number of functional connections as input variables,making the prediction results unreliable.In order to solve the problems above,this thesis uses machine learning methods to explore the relationship between verbal creativity and brain networks,proposes a method to find the optimal functional connection,and finds 13 brain functional connections that can best distinguish high and low creativity.In this thesis,we present a priori-knowledge and data-driven based approach(PDM)to find the most predictive brain functional connections.The PDM method consists of three steps: first,the rank sum test is used to select 134 functional connections that have significant differences in the high and low creative from 34716 functional connections.Then,the random forest is used to evaluate the importance of134 functional connections.The importance is sorted from high to low,and the top 20 functional connections are selected.Finally,the sequential backward selection algorithm is used to reselect the most predictive combination of function connections form the 20 functional connections,leaving 13 functional connection combinations.These 13 functional connections can effectively predict high or low creative individuals.Using 13 functional connections as input variables of the prediction model,the classification accuracy rate is 85.6% in within-database identification and 67.2% in cross-database identification,which is much higher than the classification by using the134 functional connections selected using current statistical methods.This thesis further studies the contribution of each functional connection in predicting creativity among 13 functional connection combinations.The results show that the effective classification of high and low creative individuals depends on the cooperation of 7 brain networks,which are the Fronto-parietal Task Control Network,the Sensory/somatomotor Hand Network,the Default mode Network,the Visual Network,the Cerebella Network,the Dorsal attention Network,the Cingulo-Opercular Network.The study finds that using fewer functional connections can achieve higher classification accuracy,and a single less important functional connection may contribute more to the functional connection combination.The findings of this thesis help us to better understand the neural mechanisms of the brain in human creativity. |
