Recent advances in machine learning tools have enabled substantial progress in a variety of space weather analysis, forecasting, and nowcasting techniques such as storm prediction and solar wind classification. However, extracting robust and useful features from complex time-series data can be challenging. Additionally, feature extraction methods may require careful tuning for each data product to be used as a model input, increasing their implementation complexity. We present a computationally simple, explainable, and generalizable technique for feature extraction via the automatic determination of anomalous events in time-series data products. Specifically, delay-embedding, Principal Components Analysis, and One-Class Support Vector Machines are used to identify anomalous data samples in an algorithm well-suited for deployment on spaceflight hardware. In this work, ‘anomalous data’ are intervals with distinct statistical characteristics when compared to other intervals within the time series being analyzed. Our recent work applying this technique to satellite magnetic field data demonstrates its capability to detect geophysical phenomena (e.g., discontinuities, boundaries, and waves) and local spacecraft interference (e.g., changing reaction wheel rates). Preliminary results for this data reduction technique are demonstrated against multiple data products from several spaceflight missions, and the method’s applicability to real-time machine learning and in-situ embedded hardware will be discussed.