The dependability and integrity of machine learning models are seriously threatened by attacks utilizing data poisoning. This review provides a comprehensive analysis of the literature on machine learning-based data toxicity detection. The aim is to provide a comprehensive understanding of the methods, strategies, and algorithms used to detect and thwart data poisoning attacks. In this paper, statistical analysis methodologies, model performance monitoring tools, outlier identification techniques, data sanitization techniques, and adversarial training techniques are examined. We outline evaluation metrics and data sets that are frequently used to rate detection techniques. The review also identifies current problems, roadblocks, and uncharted territory for additional research. This study is a useful resource for academics and professionals looking to improve the security and dependability of machine learning algorithms against data poisoning threats by synthesising the existing knowledge.