Aquaculture has emerged as a crucial sector in ensuring global food security, driven by the increasing world population. Smart aquaculture aims to improve efficiency, sustainability, and productivity by integrating modern tools such as sensors, artificial intelligence, and Internet of Things (IoT) into management systems. Data from IoT devices, often structured as multivariate time series, may reveal irregular patterns or anomalous features due to factors such as system failures or sensor malfunctions. Unsupervised anomaly detection techniques play a vital role in identifying abnormal behaviors within this data and are crucial for ensuring product quality and operational efficiency. In this study, three machine learning techniques (isolation forest, one-class support vector machines (OC-SVMs), and local outlier factor) and three deep learning anomaly detection techniques (autoencoder (AE), variational autoencoder (VAE), and long-short term memory AE) were analyzed using four real-world multivariate datasets collected from IoT aquaculture systems. The evaluation analysis revealed that the anomaly detection methods, OC-SVMs, and VAE demonstrated similar performance and are well-suited for detecting anomalies in multivariate aquaculture datasets.