Besides the real-time data acquired from intracranial electroencephalogram iEEG, an algorithm that identifies key features is necessary for automated diagnoses of related diseases. The power of these algorithms plays a crucial role in the accuracy of medical devices. This work reports a novel optimal feature extraction approach using wavelet transform, namely, multidepth wavelet packets (MDWPs), to accurately classify multilabeled iEEG data acquired from epileptic patients using least training data. This article also reports that the number of features employed by the algorithm is critical to the classification outcomes. In an attempt to select the optimal features, the employed algorithm obtains the MDWPs by excavating through wavelet tree down to seven levels, retaining packets at each level. Features of energy are computed, and discrete cosine transform is applied across the channels for dimensionality reduction. All the extracted features are then ranked, following which an optimal number of them is determined. This optimal feature selection allows for drawing a clear line of demarcation among all the classes, which ensures perfect classification. Contrary to the state-of-the-art models, this work, in addition to providing perfect classification results in discriminating all the five classes, also takes a smaller fraction of training data to date. The Monte Carlo scheme is employed to avoid any bias in the classification results.