Human brain is considered as complex system having different mental states e.g., rest, active or cognitive states. It is well understood fact that brain activity increases with the cognitive load. This paper describes the cognitive and resting state classification based on EEG features. Previously, most of the studies used linear features. EEG signals are non-stationary in nature and have complex dynamics which is not fully mapped by linear methods. Here, we used non-linear feature extraction methods to classify the cognitive and resting states of the human brain. Data acquisition were carried out on eight healthy participants during cognitive state i.e., IQ task and rest conditions i.e., eyes open. After preprocessing, EEG features were extracted using both linear as well as non-linear. Further, these features were passed to the classifier. Results showed that with support vector machine (SVM), we achieved 87.5% classification accuracy with linear and 92.1% classification accuracy with non-linear features.