Visual object perception is important for human's daily life. Functional brain regions on visual cortex for the familiar categories are already commonly proved by many studies. However, how do the visual cortex process the unfamiliar artificial shapes? While, there are some studies that already explored the neural representation of physical and perceptual attributes of artificial shapes; in this study, we are asking the question from another aspect: are there any representational patterns that could be used for us to discriminate both familiar categories and the artificial shapes? To answer this question, we conducted the same experiment both for artificial symbols and familiar categories using functional magnetic resonance imaging, fMRI. Then we applied machine learning models on the neural representation on visual cortex. In order to reduce the dimensionality of the features, we wrapped the features selection into the decoding model training process. Using this technique, we could manage to decode both artificial symbols and familiar categories with much less features. This feature selection technique could bring the benefits for both sides: it could improve the classification accuracy and shed a light on improving the algorithm, at the same time it gives us the selected voxels, which might contain important information related to neural representation for artificial symbols and familiar categories.