In a manufacturing industry, defected products often lead to a lot of cost. How to use quality management to improve the production process and maximize the yield is a big issue. Generally, traditional industries with relatively poor quality control methods often adjust parameters by the experience of the personnel present, which leads to many problems over time. This thesis combines data mining and regression analysis to analyze the past data of a textile factory, and use the decision tree as basic model to find the best production path. Furthermore, we design parameters by means of parameters classification. At last, statistical methods are used to verify the reliability of our model, and test whether the best production path can effectively improve the yield or not.