If you want Oracle Data Mining to treat missing data as sparse instead of missing at random or missing at random instead of sparse, transform it before building the model.
If you want missing values to be treated as sparse, but Oracle Data Mining would interpret them as missing at random, you can use a SQL function like NVL to replace the nulls with a value such as "NA". Oracle Data Mining does not perform missing value treatment when there is a specified value. See Oracle Database SQL Language Reference for details about the NVL function.
If you want missing nested attributes to be treated as missing at random, you can transform the nested rows into physical attributes in separate columns — as long as the case table stays within the 1000 column limitation imposed by the Database. Fill in all of the possible attribute names, and specify them as null. Alternatively, you could insert rows in the nested column for all the items that are not present and assign a value such as the mean or mode to each one.