The human brain is like a network composed of different areas. Each of these areas has its function, as well, they interchange their data continuously. Nowadays, modeling the brain network, and achieving the states it encounters during various behaviors are essential. To this aim, it is required to have some knowledge about both structure and function of the brain areas against each other. Each function in the brain is the result of a circuit of some connected areas of the brain. Cognitive disorders occur when there is some kind of abnormality in these areas' activity and the level of connectivity between them. In other words, brain disorders affect the way the brain connectivity. In this research, the focus is on the effects of autism disorder (ASD) on these connectivities. Autism is a disorder that affects the brain's growth in childhood and is characterized by symptoms like difficulty in social interactions. Many reports are illustrating the difference in functional/effective connectivity between autism and healthy people. These investigations are mainly based on fMRI images of the brain during task or rest states. Standard methods for evaluating the connectivities are based on seed-based and ROI-based correlation between different brain areas. ROIs have defined by clustering methods like independent component analysis (ICA). Although these researches declared many facts about autism and its mechanisms, none can be sufficiently accurate to early diagnosis of autism in clinical usage. Assuming functional connectivity constant during fMRI imaging (referred to as static functional connectivity) is the main drawback of this researches. In this project, we consider temporal changes of functional connectivity (dynamic functional connectivity) to find a more reliable pattern discriminating against the healthy and autistic brain. This idea leads us to a significant amount of connectivity parameters, i.e., a sequence of connectivity matrices. Different methods, including clustring of connectivity matrices, using kmeans, cmeans, and ICA methods, extraction of information of these matrices as a feature vector using different entropy methods are proposed. We evaluate our method by ABIDE1 database, including 573 healthy and 539 ASD people. Finally, by combining correlation methods and feature selection methods introduced in this study, the accuracy of autism diagnosis is improved. Finally, we conclude that the cmean algorithm is more accurate than the other algorithms used when all laboratories are Key words- functional connectivity, fuzzy cluster,entropy, autism, Magnetic resonance imaging, dir="RTL"