At present Brain Computer Interface (BCI) technology plays an important role in the field of biomedical. BCI can be used to identify human stress based on various emotion states - happy, sad, disgust, fear. Various techniques such as Neural networks (ANN), statistical methods, autoregressive model (AM), mixture of densities approach, independent component analysis (ICA), time-frequency analysis (TFA), bayes quadratic, hidden markov model (HMM) and linear discriminate analysis (LDA) have been used for EEG signal analysis. There are some issues which are being focused on while extracting features are -handling irrelevant and redundant features require more computation for decomposition of EEG signals. And the paper is considering following issues while carrying out classification - inability in processing complicated set of data, low performance when training data set is huge. This paper focuses on addressing above mentioned issues and appropriately analyzes various EEG signals which would in turn help us to improve the process of feature extraction and improve the accuracy in classification. In addition to acknowledging above problems, this paper proposes a framework which would be helpful in identifying human stress level and as a result, differentiate a normal or stressed person/subject.