Federated learning (FL) is a privacy-preserving machine learning approach that enables multiple parties to collaboratively train a shared model without sharing their raw data. It addresses the challenge of data privacy in distributed environments by allowing data to remain decentralized while still benefiting from the collective knowledge. However, due to this collaborative training of a shared model, it has been known that FL is susceptible to various poisoning attacks where a participant intentionally submits manipulated data or maliciously alters their model updates to compromise the integrity and accuracy of the federated learning model. This chapter provides a comprehensive overview of poisoning attacks in FL and explores recently developed defense methods. Next, it focuses on a state-of-the-art defense algorithm called LoMar (Local Malicious Factor), which utilizes a two-phase approach to detect and mitigate attacks. In phase I, LoMar scores model updates based on the relative distribution over neighboring participants using kernel density estimation. In phase II, an optimal threshold is approximated to distinguish between malicious and clean updates. Extensive experiments on real-world datasets are conducted to compare existing defense mechanisms in protecting FL systems against data and model poisoning attacks.