Accurate prediction models for air pollution are crucial for forecast and health alarm to local inhabitants. Discrete wavelet transform (DWT) was employed to decompose a series of air pollutant levels, followed by modeling using support vector machines (SVMs) in recent literature. This combination of techniques was reported to produce a more accurate prediction model for air pollutants by investigating different levels of variations. However, DWT demands significant model complexity, namely, the training time and the model size of the prediction model. In this study, a new method called variation-oriented filtering (VF) was proposed to remove the data with low variation which can be considered as noise. Using VF, the size of training data and the number of support vector can be reduced, which can reduce training time, computational complex and model size of prediction model; hence the efficiency of prediction model can be improved. Simultaneously, the noise in the series of air pollutant levels can be filtering by using VF, which can improve the accuracy of prediction model. The SPM (suspended particulate matter) level in Macau was selected as a test case. Experimental results showed that VF can effectively and efficiently reduce the model complexity with improvement in predictive accuracy, as compared with DWT.