This is just one example of creating an instance in-memory. I think this way is readable and provides enough flexibility on attribute types (numeric, nominal, string).

CreatingWekaInstance.java

Instances dataSet = // ...
Attribute stringAttribute = // ...
Attribute nominalAttribute = // ...

int numberOfAttributes = 3;
Instance instance = new SparseInstance(numberOfAttributes);
instance.setDataset(dataSet);

instance.setValue(numericAttribute, 8.8);

/* 
 * The way Weka stores data is a bit low-level. For any
 * CRUD operations, we have to care about the index number
 * of nominal and string values.
 * As a comparison, CRUD operations in SQL or Search
 * Engine Database are already abstracted.
 */

instance.setValue(nominalAttribute, dataSet.attribute(
  "gender").indexOfValue("male"));

String userIdAsString = "8379";
stringAttribute.addStringValue(userIdAsString);
instance.setValue(stringAttribute, userIdAsString);

dataSet.add(instance);