An Empirical Analysis Of Farmers' Income Based On Rough Set And Support Vector Regression

China is an agricultural oriented country, farmers is an important part of our population. The slowly increasing of farmers'income is restricting economic development in china. Many experts and scholars both of the foreign and domestic have discussed and studied this problem.Based on previous researches, we selected basic situation of net income per capital of rural residents between 1983 and 2006 from the People's Republic Bureau of Statistics Web site as the original data. Eight indexes were selected after investigated and analyzed, they are net income per capital of rural households, wage income, net income of family business, agricultural income, total income of forestry, animal husbandry, fisheries, total income of industry, construction, total income of others and transfer income.Firstly, we fit net income per capita of rural households with income of other approaches by kinds of function (the selected functions are linear function, logarithmic function, inverse function, quadratic function, cubic function, compound function, power function, Sigmoid function, growth function, exponential function, logistic function), investigate the intrinsic link between net income per capita of rural households and every components of the indexes.Secondly, in order to find the relationship between net income per capita of rural households and its components, stepwise linear regression was made between net income per capita of rural households and income of other approaches, the data are preliminary by integrated analyze. Then rough set theory and method was used to discrete data of simplified (Equal Interval Width) and the attribute reduction (taken income per capita of rural households as decision property, and the remaining index as condition property), the dimension of sample data is effectively reduced by the above two-step pretreatment.Finally, we put forward support vector regression model. The data of twenty years from 1983 to 2002 was taken as training samples, and data of four years from 2003 to 2006 was taken as the test samples, we select radial basis function as kernel function,ε-insensitive function was taken as loss function, set C=200,ε=0.01, Satisfactory result is obtained by the calculation of mathematics software.