Classification, which is one of the most powerful approaches to filter valuable information from big data, is a typical supervised learning method of machine learning. In many real applications, the collected data may contain many redundant features and have some missing entries. If a classifier is learned directly on such data, it cannot obtain a satisfactory classification performance. In this paper, we propose an ensemble classification framework based on latent factor analysis (ECF-LFA). Its main idea includes two parts: 1) employing the latent factor analysis (LFA) to extract the latent factors (LFs) from original data, which can avoid the influence of redundant features and handle the data with many missing entries, and 2) using these extracted LFs as the input for base classifiers to conduct the ensemble learning, which can boost a base classifier's classification accuracy. Experimental results on four benchmark datasets and three well-known classification algorithms verify that ECF-LFA can effectively improve a classifier's performances.