疟疾预测的遗传规划方法与应用——以安徽省县（市）疟疾发病率为例

疟疾是世界上最严重的一种寄生虫疾病,安徽省是典型的中纬度疟疾高发区域之一。本文以安徽省县级行政单元统计的疟疾发病率为例,从遥感监测数据中获取疟疾潜在驱动因素的数据,使用遗传规划方法建立遥感监测的环境因素与疟疾发病率之间的关系,从而预测疟疾发病率的空间分布,并分析预测结果、评价模型精度。结果表明,遗传规划方法预测的疟疾发病的精度（训练数据的预测R² = 0.558,检验数据R² = 0.429）较线性逐步回归方法的预测精度（训练数据的预测R² = 0.470,检验数据R² = 0.408）有所提高。遗传规划方法有利于提高预测疟疾发病率空间分布的精度。其为使用遥感监测数据预测疟疾的空间分布和变化的科学研究提供依据。

Application of Genetic Programming on Predicting and Mapping Malaria in Anhui Province

This paper delineates the relationship between remote sensing monitoring indexes and malaria incidences using genetic programming (GP) method based on factors derived from remote sensing data. Thus, the spatial distribution of malaria incidence is predicted, the prediction results are analyzed, and the modeling precision is evaluated. Malaria is considered to be the severest parasite disease and Anhui Province is one of the typical mid-latitude areas coping with high malaria risk. This paper studies the issue of predicting malaria spatial distribution using GP method, as GP is a striking optimization method which has the capability of exploring a proper solution for sophisticated issues through evolutionary algorithms. And this process is further explained with an example adopting the monthly average malaria incidences in each county of Anhui Province from 2004 to 2010. Also, remote sensing data is regarded to be the main source of factors, considering its large spatial scale and fast data acquisition, and that various meteorological and environmental indexes, could be converted from remote sensing data. These factors include remote sensing indexes, such as normalized difference vegetation index (NDVI) and land surface temperature (LST), plus natural attribute (elevation) and social attributes (population, immigrant and GDP data) in the county level. Results demonstrate that NDVI and LST have influences of two months’ and one month’s lag respectively. Compared with the result of linear regression (R² = 0.470 for training data and R² = 0.408 for test data), the predicting precision is improved using GP method (R² = 0.558 for training data and R² = 0.429 for test data), which is benefited from illustrating the non-linear relation between remote sensing indexes and malaria incidences. GP method contributes to increase the precision of predicting the spatial distribution of malaria incidence. Conclusively, this paper provides a basis for future scientific research on predicting spatial distribution and mapping malaria using remote sensing data.