Electroencephalogram (EEG)-based emotion identification is important for human–computer interface (HCI) applications. Neurological alterations in the brain produce EEG signals. Decomposing EEG data into five frequency bands (delta, theta, alpha, beta, and gamma) is useful for effective analysis. EEG signal rhythms are useful in isolating noise from EEG and identifying useful neurological patterns. This paper proposes a novel framework for decomposing EEG signals into rhythms using optimized infinite impulse response (IIR) bandpass filters. The crayfish optimization algorithm (COA) optimizes the filter’s parameters, such as filter order and cutoff frequencies. The COA algorithm finds the optimum values for filter parameters by minimizing the mean squared error between the original and reconstructed signal derived from rhythms. Multiple statistical characteristics, such as Hjorth Mobility, Hjorth Complexity, Hjorth Activity, and Shannon Entropy, are derived from the EEG rhythms. These statistical features are input to different versions of Support Vector Machines (SVM), Trees, Discriminant, Naive Bayes, k-NN, Ensemble, and Neural Networks. The best classification accuracy of 94.4% and 93.0% is observed with cubic and quadratic SVM better than the existing methods.