Abstract: Missing information, indefinite information as well as scarcity of information truly exist due to the complexity of the real world. Relational database, as an important tool to express the real world, uses null value to express the missing information. Focusing on estimation of null values in relational databases, the paper proposes a multi-null value estimation method based on multi-table relationship information. First, it arranges the sequence of estimating null values of each attribute based on the principle of minimizing the bias that is brought in. Then it estimates null values of each attribute based on the information of the basic table. After that it brings in multi-table relationship information when the forecast error exceeds a threshold value. The schema to be brought in information depends on the relationship between the basic table and other related tables. In that case the proposed method can improve the accuracy of forecasting null values. The experiment results show that the proposed method is of relatively high accuracy.

Key words: relational database, null value, fuzzy clustering, regression coefficient