Neuromarketing is the application of neuropsychology in marketing research to study the sensorimotor behavior of consumers such as cognitive and emotional responses to marketing stimuli with the help of new technologies. The performance of priority detection systems based on brain signals depends on the appropriate selection of feature extraction methods and machine learning algorithm. At first, EEG signals were pre-processed by low-pass and Savitzky Golay filters, and the features of discrete wavelet transform (DWT) and Lampel-Ziv complexity (LZC) were extracted from them. After that the features are normalized and divided into training and testing. Later on, the normalized features are given to a four-layer deep neural network (DNN) to predict the results of the training. Finally, the proposed model is ready to use. To evaluate the performance of the proposed model, parameters of precision, recall, and accuracy have been considered. The results show that a precision of 82%, recall of 87.5%, and accuracy of 87.5% for distinguishing the two categories of liking and disliking have been obtained in an average of five runs. In this study, the effect of advertising on the brain activity of consumers was investigated by analyzing EEG signals. Experimental results of the proposed model show that studies in this field can change and improve marketing strategies to improve producer performance and consumer satisfaction, leading to a mutual benefit.IntroductionNeuromarketing is the study of brain reactions using medical technologies in response to marketing stimuli (Amin et al., 2020). By analyzing the collected information, companies try to find the reasons for consumer decisions to buy goods. They also want to understand which brain regions become more active during decision-making.Neuromarketing researchers assume that most consumer decisions are made unconsciously and in a fraction of a second. They also believe that the choices and decisions consumers make are often based on emotions instead of product comparison. Therefore, consumers’ emotions influence decision-making. On the other hand, consumer feelings can be highly influenced by internal and external factors. Detecting the consumer’s emotional state reveals his/her real preferences (Aldayel et al., 2020). The feeling and understanding caused by advertising make the consumer buy a particular product.Neuromarketing studies are done through recording or analyzing biometric data, or through a combination of them such as electroencephalography (EEG), facial expression, movement pattern recognition, eye tracking, and galvanic skin response (GSR). A consumer preference recognition system based on EEG signals helps to understand consumer behavior and discover how a person makes a purchase decision. Understanding this, helps marketers and organizations to increase customer satisfaction, positive customer experiences, consumer loyalty, and revenue (Aldayel et al., 2020). Materials and MethodsIn this article, the EEG signals of 25 participants were used while viewing 14 different internet products of three different types. Such signals were recorded through 14 channels at different head surface areas with the international 10-20 system including AF3, F7, F3, FC5, T7, P7, O1, O2, P8, T8, FC6, F4, F8, and AF4. Common mode sense (CMS) and driven right leg (DRL) reference electrodes are placed at P3 and P4 positions above the ear. First, the EEG signals are pre-processed by band pass filter and Savitzky Golay filter. Then, the discrete wavelet transform (DWT) and Lampel-Ziv complexity (LZC) features are extracted from them. After that, normalized features are divided into two parts of training and testing. In the proposed model, the four-layer deep neural network (DNN) classifier is used to predict and distinguish two categories of liking and disliking. The input layer of DNN is not recognized as a layer in the network and is known as zero layer. Hidden layers include units with rectified linear unit (ReLu) (Kingma & Ba, 2014; He et al., 2015). The output is configured as a softmax layer with a binary cross-entropy cost function. Each hidden layer consists of 60% of its previous layer; therefore, the first hidden layer contains 1800 units, the second hidden layer contains 1080 units, and the third hidden layer contains 648 units. The dimension of the output layer is equal to two, because it corresponds to the number of units required to distinguish the categories of liking and disliking. ResultsThe simulations have been performed by MATLAB 2014a software. The training set included 80% of extracted features while the testing set included 20% of extracted features. Various evaluation criteria such as precision, recall, and accuracy have been used to measure the performance of the classifiers. The values of statistical parameters after averaging five times to distinguish two categories of liking and disliking have been found to have 82% precision, 87.5% recall, and 87.5% accuracy. DiscussionVarious evaluation criteria such as precision, recall, and accuracy have been used to measure the performance of the classifier. The proposed model was able to distinguish the two categories of liking and disliking with 87.5% accuracy, which shows an increase of 0.5% compared to the reference (Aldayel et al., 2021). Also, by comparing the results with references (Aldayel et al., 2021; Yadava et al., 2017), it is clear that the LZC feature extraction method has an effective role in predicting the proposed model for separating the two categories of liking and disliking. The results show that the proposed method can provide a complementary solution to traditional measures of predicting products success in the market. Also, the proposed method can be used in the development of market strategies and research by predicting the success of the market through expanding the existing models. This method helps advertisers understand which of their advertisements are effective and how they should plan their advertising strategy. Finally, by comparing the results with previous studies, the combined feature extraction method of LZC and DWT with the use of DNN classification has an effective role in predicting the proposed model for separating the two categories of liking and disliking. *