Probabilistic Graphical Models is a powerful framework for representation and inference in probability distributions with many random variables. Sequence labeling is one of the most challenging problems in the field of pattern recognition. In this problem, we want to predict an output vector given a sequence of observations. It has many applications such as part of speech tagging, handwritten text recognition, speech recognition, protein secondary structure prediction and human action recognition. So far, a lot of probabilistic models such as HMMs, MEMMs, CRFs, HCRF and CNFs have been used to solve this problem efficiently and accurately. In models like CRF the distribution over output variables is a log-linear function of observations. In practice, the relation between inputs and outputs is highly non-linear. In this thesis we propose a model to assign a single label to a sequence of observations which capture the non-linearity between inputs and outputs by using a layer of ANFIS networks. We evaluate the proposed model on the task of human action recognition using skeleton data and show that our model achieves better results than models like HCRF and CNF. Key Words Probabilistic Graphical Models, Conditional Random Fields, Hidden Conditional Random Fields, Conditional Neural Fields, Human Action Recognition