Urinalysis remains one of the most commonly performed tests in clinical practice. Laboratory work can be greatly relieved by automated analyzing techniques. However, noisy and imbalanced urine samples make automatically identifying and classifying urine-related diseases become very difficult. This paper proposed hybrid sampling-based ensemble learning strategies by improving training data and classification performance. Having compared the effectiveness of several learning classifiers and data processing techniques, the experiments showed that the suggesting methods provided better classification accuracy than other approaches.