Tackling air pollution is a challenging problem of great importance in environmental studies. Most previous studies tackled this problem primarily by developing prediction models using statistical, machine learning, and deep learning techniques. Unfortunately, the successful industrial adoption of these studies has been limited by their inability to discover the valuable information hidden within the predicted data. This paper proposes a novel binary neuro-symbolic artificial intelligence framework to discover the valuable information hidden within the predicted air pollution data of multiple sensors in a network. The neural component of our framework involves model building and predicting the air pollution for each sensor in a network. The symbolic component of our framework involves data transformation of predicted air pollution data of multiple sensors into a transactional database using novel symbolic rules and discovering frequently occurring pollution patterns in the predicted data. Experimental results on six years of real-world air pollution data gathered throughout Japan show that the proposed framework is efficient and finds valuable information.