Detection of Causative Attack and Prevention Using CAP Algorithm on Training Datasets

Abstract

Machine learning is the scientific study of algorithms, which has been widely used for making automated decisions. The attackers change the training datasets by using their knowledge, it cause impulses to implement the malicious results and models. Causative attack in adversarial machine learning explores certain security threat against carefully executed poisonous data points into the training datasets. This type of attacks are caused when the malicious data gets injected on training data to efficiently train the model. Defense techniques have leveraged robust training datasets and prevents accuracy on evaluating the machine learning algorithms. The novel algorithm CAP is explained to substitute trusted data instead of the untrusted data, which improves the reliability of machine learning algorithms.