Epileptic seizures are unpredictable in nature and its quick detection is important for immediate treatment of patients. In last few decades researchers have proposed different algorithms for onset and offset detection of seizure using Electroencephalogram (EEG) signals.

In this paper, a combined approach for onset and offset detection is proposed using Triadic wavelet decomposition based features. Standard deviation, variance and higher order moments, extracted as significant features to represent different EEG activities.

Classification between seizure and non-seizure EEG was carried out using linear discriminant analysis (LDA) and k-nearest neighbour (KNN) classifiers. The method was tested using two benchmark EEG datasets in the field of seizure detection.

CHBMIT EEG dataset was used for evaluating the performance of proposed seizure onset and offset detection method.

Further for testing the robustness of the algorithm, the effect of the signal-to-noise ratio on the detection accuracy has been also investigated using Bonn University EEG dataset.

The seizure onset and offset detection method yielded classification accuracy, specificity and sensitivity of 99.45%, 99.62% and 98.36% respectively with 6.3 s onset and −1.17 s offset latency using KNN classifier.

The seizure detection method using Bonn University EEG dataset got classification accuracy of 92% when SNR = 5 dB, 94% when SNR = 10 dB, and 96% when SNR = 20 dB, while it also yielded 96% accuracy for noiseless EEG.

The present study focuses on detection of seizure onset and offset rather than only seizure detection. The major contribution of this work is that the novel triadic wavelet transform based method is developed for the analysis of EEG signals. The results show improvement over other existing dyadic wavelet based Triadic techniques.