Bioinformatics has been one of the most fast developing fields in scientific research. As an important tool in bioinformatics, microarray technology was developed in a great
need. It provides a remarkable methodology to monitor the expression level of a large amount of genes. This technology has created a number of new opportunities to study human disease. However, the resulting mass of data, often as millions of measurements, can be overwhelming for human comprehension and pose great challenges on data pre-processing and analysis methodology. One of the major difficulties with microarray data is the presence of missing values that are commonly due to a number of diverse reasons. These include insufficient image resolution, dust or scratches on the slide or experimental error during the laboratory process. Generally, missing of data around 1-10% would affect up to 95% of the genes in any microarray experiment [3]. Of course, one
possible solution to recover the dataset with missing values is to repeat the experiment, which is inefficient and sometimes infeasible. Henceforth, the missing values in the original data need to be systematically predicted.