Battery field data collection enables car manufacturers to monitor battery aging throughout the entire life cycle. However, to fully utilize the potential of this data, it is essential to identify anomalous aging estimates, as existing methods have not yet demonstrated the ability to effectively distinguish outliers from nominal batteries in real-world applications. In this work, we provide an unsupervised framework based on the big data of battery field readouts using statistical learning for anomaly detection. Our dataset includes 12 million readouts from 600 thousand vehicles. Our method effectively clusters estimated aging values to group batteries with similar aging conditions, leveraging a statistical learning model to detect anomalies within these clusters. To validate our model, we use real-world data from a corrupted prototype software. We successfully identify 98 % of the vehicles with irregular battery aging. Our framework offers robust identification of anomalous estimated aging values, facilitating cost-sensitive decisions such as battery replacements, ultimately optimizing overall battery operation strategies.