This study investigates the use of machine learning for anomaly identification in satellite telemetry. It seeks to answer important research problems, such as how well different machine learning models identify abnormalities in extremely unbalanced telemetry data and how accuracy may be improved by using ensemble approaches. This study assesses models including ARIMA, RNNs, LSTMs, Isolation Forests, and K-means clustering using a publicly available dataset. To enhance detection performance, a unique ensemble approach that integrates several models is suggested. With a focus on accuracy, recall, and computing efficiency, the results show each model's advantages and disadvantages. The results can be used in various fields that need early anomaly detection in time-series data and offer insightful information for enhancing satellite health monitoring.