Olfaction Recognition by EEG Analysis Using Differential Evolution Induced Hopfield Neural Net Anuradha Saha, Amit Konar, Pratyusha Rakshit ETCE Department Jadavpur University, Kolkata, India. anuradha.nsec@gmail.com, konaramit@yahoo.co.in, pratyushar1@gmail.com. Anca L. Ralescu Computer Science School of Computing Sciences and Informatics College of Engineering University of Cincinnati, Cincinnati, USA anca.ralescu@uc.edu. Atulya K. Nagar Department of Math and Computer Science Liverpool Hope University, Liverpool, UK. nagara@hope.ac.uk. Abstract— The paper proposes a novel approach to recognize smell stimuli from the electroencephalogram (EEG) signals acquired during the period of inhalation. The main contribution of the paper lies in feature selection by an evolutionary algorithm and pattern classification by Differential Evolution induced Hopfield neural network. One additional merit of the work lies in data point reduction by Principal component analysis. Experiments undertaken on 25 subjects with 10 smell stimuli indicate that the proposed scheme of feature selection, data point reduction and classification outperforms the traditional approach by a wide margin. Experimental results confirm that the smell stimuli excites the pre frontal lobe of the human brain and is responsible for a special type of brain rhythms (EEG signal) in alpha-band, theta-band and delta-band. Keywords— olfaction; electroencephalogram signal; differential evolution; principal component analysis; Hopfield neural network. I. INTRODUCTION Olfaction refers to the sense of smell involved in identification of foods, fragrances, and chemical stimuli. This paper attempts to examine the effect of olfaction on electroencephalogram (EEG) [1] signals. Apparently, odor encoding and perception involves memory cells, namely long- term memory, to retrieve and match odor of known stimuli [2]. Participation of neurons in smell stimuli processing vary widely depending on the nature and complexity of odor detection. Here, we analyze the EEG pattern during different odorant inhalation followed by the theoretical guidelines and distinguish the different odorants with the basis of the extracted features regarding the stimuli. Signal modality is an important issue in EEG research. In the present context of olfaction recognition, we used SCP as the modality of EEG signal. Slow cortical potential, as found in previous study [3], may be defined as gradual changes of potential spotted in the membrane of dendrites that lasts from 300ms up to seconds. SCP originates due to feedback and positive reinforcement mechanism [3]. Here, the positive potential shift of SCP is generated during smell recognition state. Selection of classifier has been guided by two issues here: i) fewer number of classes and ii) possibility of misclassification by the source noise in smell stimuli. Hopfield neural net [4] here is an ideal choice for the aforesaid reasons. First, the energy function of the Hopfield net can be tailored to have fewer minima corresponding to the stable state of the Hopfield dynamics. Second, noisy stimuli ultimately settle down to the desired optima as the energy surface contains fewer and widely spaced optima. In this paper, we optimally select the weight matrix by minimization of the energy function of the Hopfield dynamics using Differential Evolution (DE) algorithm [5]. The evolutionary feature selection algorithm constructed for the present application aims at identifying the essential features of the data points so as to satisfy two criteria jointly. First, each selected feature i should have a narrow valuation space for all the data points lying under each class. Second, for each selected feature i, the class mean of the feature for any two classes should differ by a wide margin. An objective function is created to jointly optimize the above two objectives, and a DE algorithm has been employed to determine the essential features that jointly optimize the given objective function. Very few works on EEG analysis for smell stimuli classification are found in the literature [6]-[9]. One of the pioneering works in this regard is due to Boeijinga and Lopes da Silva [6]. They examined the possible correlation between olfactory stimulus and the nonlinearity in EEG. Inspired this work, Harada et al. [7] attempted to identify possible correlation between olfactory stimulus and brain functionality induced by EEG. Almost at the same time, Nicolas et al. [8] studied the effect of odor at several consumers’ behavior. In [9] also, EEG signals are extracted to classify personal preferences of olfactory stimulus by using factor analysis. Henkin and Levy [7] examined the correlation of pleasant (unpleasant) stimuli with EEG signal taken from left (right) hemisphere. In fact, by applying theories and computational intelligence, researchers developed a technical device, known as ‘electronic nose’ [10], which can successfully detect some significant non-odorable gases. Unfortunately, none of the above works directly dealt with classification of source of odors from EEG signals. This paper proposes a novel approach to classify 10 different sources of odors using EEG signals with special emphasis in i) exclusive feature selection, ii) data point reduction and iii) classification of smell stimuli.