Given a new class imbalanced dataset D and limited computational resources, the challenge arises of selecting promising class imbalanced learning (CIL) pipelines that include resampling methods, classification models, and their corresponding hyperparameters. To address this challenge, we study Zero-shot Automated Machine Learning and propose a new approach aiming at class imbalanced data, called Zero-shot Automated Class Imbalance Learning (ZAutoCIL). ZAutoCIL employs domain-independent meta-learning to develop a zero-shot surrogate model for automated class imbalanced learning. This model aims to recommend effective CIL pipelines for new unseen imbalanced datasets without requiring additional search. Specifically, we meta-train a two-tower model to serve as the surrogate model, adapted from recommender systems, using a pairwise ranking loss on the meta-dataset gained from collecting performance data across a wide range of CIL pipelines and a comprehensive repository of class imbalance datasets. We perform extensive experiments on 100 datasets grouped in 4 parts based on their imbalance ratio. The experimental results demonstrate the efficacy of our approach in automating the recommendation of CIL pipelines given any target imbalanced datasets.