A Neuro-C4.5 Paradigm for Audio Steganogram Detection Based on Asymmetric Cost of False Errors

This paper proposes an Audio Quality Metrics (AQM) based steg analysis for detecting audio steganograms. Primarily two contributions are made to achieve superior detection rates of the proposed steganalyser. First, an effective learning algorithm is employed for audio steg analysis, neuro C4.5, which possesses good comprehensibility and generalization ability. Second, the asymmetric costs of false positive and negative errors are investigated to enhance the steganalyser´s performance. Empirical results show that the neuro-C4.5 model, designed based on the asymmetric costs of false negative and false positive errors proves to be effective.