Fabrication is when you invent evidence and claim it to be real. It could be results from an experiment that you did not do or an interview that you did not conduct.
Fabrication is not as uncommon as you might think. A survey among American health researchers conducted in 2005 showed that three out of every thousand respondents had fabricated data within the past three years.
Fabrication is considered misconduct, and is likely to be heavily sanctioned if detected.