In our last research, we proposed the AndroBank system for creating Datasets of guaranteed quality. In this paper, we draw attention to the fact that using contrasting datasets can bias AI results. We conducted an experiment using the AndroBank tool to create a Contrasting dataset. This experiment fully supports our claims about the possible degradation of the detection results of machine learning models. The usage of contrasting datasets in academic research is quite common, as it results from the most basic concept of gathering samples. This is closely related to the „Delayed Interception“ phenomenon, which has been defined in the paper.
Highlights:
- Definition: Dataset of guaranteed quality
- Definition: Delayed Interception
- Definition: API Milestones
- Experiment: Impact of API Levels
Complete list of used samples in Contrasting dataset: 01_SDK_PUBLISH_out