Investigating machine learning attacks on financial time series models

Machine learning and Artificial Intelligence (AI) already support human decision-making complement professional roles, are expected in the future to be sufficiently trusted make autonomous decisions. To trust AI systems with such tasks, a high degree of confidence their behaviour is needed. However, can drastically different decisions if input data modified, way that would imperceptible humans. The field Adversarial Learning studies how this feature could exploited by an attacker countermeasures defend against them. This work examines Fast Gradient Signed Method (FGSM) attack, novel Single Value attack Label Flip on trending architecture, namely 1-Dimensional Convolutional Neural Network model used for time series classification. results show architecture was susceptible these attacks that, face, classifier accuracy significantly impacted.