Support Vector Machine, Naïve Baye’s, and Recurrent Neural Network to Detect Data Poisoning Attacks on Dataset

Machine learning is a subset of artificial intelligence, that has revolutionalized the world in recent days. The machine learning model will be trained and tested on the collection of huge data, a large portion is for training and others for testing. data has been collected from trusted or untrusted sources, and results will be predicted using different algorithms. People are getting used to these predictions, are these predictions secure, trustful, or guaranteed? Well, that will be decided by the quality of the datasets. Although the datasets can be poisoned easily. Here the proposed work is providing a solution to check the quality of the dataset used for machine learning algorithms. The basis of the research is to detect poisoning attacks in the data set and to determine the most accurate estimate of the detection of poisoning attacks. comparing accuracies of the supervised machine learning algorithms Support Vector Machine(SVM), Naive Baye’s Classifier (NBC), and deep learning algorithms Recurrent Neural Network(RNN).The evaluation report was submitted for the results obtained using the rating report. The classification report considers performance measures the Precision, Accuracy, and F1 score of each algorithm.

Introduction

Machine learning is the process of separating information from data so that it may be transformed into useful goals. Moreover, machine learning along with computer vision has augmented domains such as medical diagnostics, scientific research, and statistical data analysis. Also, it is being used in Email intelligence, social network, banking and finance, and personal smart assistant. People are influenced by how machine learning and artificial intelligence have changed our life by making them easy. Is it correct? In daily life directly or indirectly, the machine learning models are ruling humans, from what to eat, where to go and what to read too. But from where do these predictions or suggestions come from? We design machine learning models to do this all, train, and test it on large datasets. The datasets are the collection of data collected from various sources. But the threat for the machine learning algorithms is the dataset has the chance of poisoning attacks. The poisoning attacks mean adding bad data to the dataset used to train and test or misleading the training models. Data poisoning attacks have the potential to reduce the accuracy of algorithms, which could be detrimental. The machine learning models expect to have the highest accuracy possible. If data set with poisoning attacks will be used to train the healthcare diagnosis algorithms, banking, financing, evolution, and assessments, other decision-sensitive areas, might be harmful financially, physically as well as mentally. To anticipate and deal with poisoning attacks detection must be done. Detection of poisoning attacks on datasets is beneficial to distinguish between true data and poisoned data. The goal is to identify the deep learning method Recurrent Neural Network (RNN), the Naive Baye’s Classifier (NBC), and the support vector machine (SVM) as the models with the highest accuracy for detecting data poisoning attacks.