Brain computer interface (BCIs) aims to communicate with external devices and also recognize human activities through brain signals. However, existing algorithms have some drawbacks, such as poor resolution, high-frequency noise, risk, and low accuracy. Various machine learning algorithms have been used to recognize human activities, but it is more complex to recognize the activities of humans. In order to solve these issues BCIs system based on modified DNN is proposed. Using a modified DNN algorithm, all human activities are recognized through the brain signals. The proposed method contains two phases such as training and testing the data. First, the Brain signal is taken as input, and then the next phase is pre-processing. The signal is pre-processed using a bandpass filter to remove noise present in the signal. Then the next phase is feature extraction, for that features are extracted from the pre-processed signal using Hybrid Adaptive Filtering (HAF) and Higher Order Crossing (HOC) analysis. In order to evaluate the HAF-HOC performance on a user-independent basis, a set of brain signals for each human activity was analyzed and pre-processed. The next phase is classification. For classifying the brain signal, a modified DNN algorithm is used. The Sailfish Optimization technique is used to train the neural network and also used to optimize the weights of the DNN classifier. The results show that the proposed method can obtain 98% accuracy and 0.03 Error, and 10% False Positive Rate. An existing system such as SVM, KNN, RF, and NB reached 86%, 90%, 83% and 70% accuracy, respectively, which is lower than our proposed method.