Boosting-based Autoencoder

• Adapted boosting mechanisms on simple autoencoders for outlier detection purposes specialised in discarding outliers in each boosting iteration.

• Surpassed state-of-the-art and classic anomaly detection strategies achieving an average AUCPR score of 46.2%.

• Examined the relationship between the data distribution and the correct prediction rate of BAE with respect to the other approaches proposed in the literature.

• Performed exhaustive experiments on the performances of BAE and their superiority on all test-beds according to statistical significance methods.

• Collaborated in writing and formalising the methodology’s workflow by relying on mathematical notation of algebra and first order logic.

Bardh Prenkaj
Bardh Prenkaj
Computer Scientist, PhD

It’s okay to stalk for professional reasons!