|Boy Nurtjahyo Moch.||- Department of Industrial Engineering. Faculty of Engineering, Universitas Indonesia
|Komarudin Komarudin||Department of Industrial Engineering, Faculty of Engineering, Universitas Indonesia|
|Maulana Senjaya Susilo||Department of Industrial Engineering, Faculty of Engineering, Universitas Indonesia|
Fixation points, as the stopping location of eye movements, can be extracted to generate valuable information about a picture or an object. This information is valuable as it enables the identification of the area/part of the picture that attracts people’s attention, which can be used as a consideration when making decisions in the future, for example in marketing. For this reason, in this study, a Neural Network (NN) model was developed to predict the fixation points of a picture. Specifically, the authors experimented with various transfer and training functions in the NN in order to determine which causes the fewest errors. The results show that the method used is applicable in practice since it produces MAPE (Mean Absolute Percent Error) of around 13–15% and MSE (Mean Squared Error) of 0.9–1.1%.
Eye tracking; Fixation points; Neural network; MAPE; MSE
From several analyses that have been performed, there are several conclusions. In terms of the accuracy of the prediction model, as measured by the smallest error value, the best combination of functions is purelin-purelin, trainscg. The combination of these functions ranks first in MSE and second in MAPE calculations. In addition, in terms of the computing performance of the prediction model, as measured by the smallest number of iterations and the shortest training duration, the best combination of functions is purelin-purelin and trainbfg. Moreover, the trainscg training function produces a smaller range of MAPE values than traingdx or trainbfg. Lastly, the trainbfg training function involves a shorter training duration and a smaller number of iterations than traingdx or trainscg.
For future research, several future works can be proposed. Instead of viewing a picture, respondents may be more attracted to seeing human faces. Therefore, future work could use pictures that include a human face as the research object. In this situation, the testing would be more accurate if additional tools were employed, for example the Viola Jones face detector and the Felzenszwalb person detector.
In addition, the combination of the training and testing carried out in this research is limited only to a combination of inter-layer transfer and training functions. Therefore, more research should be performed to investigate different combinations of numbers of layers and network types.
This study was financially supported by Hibah PITTA 2017 from the Directorate of Research and Community Engagement, Universitas Indonesia.
Bang, H. and Wojdynski, B.W., 2015. Tracking Users' Visual Attention and Responses to Personalized Advertising based on Task Cognitive Demand. Computers in Human Behavior, Vol. 55, pp. 867–876.
Bhardwaj, A., Tiwari, A., Bhardwaj, H. and Bhardwaj, A. 2016. A Genetically Optimized Neural Network Model for Multi-class Classification. Expert Systems With Applications, Vol. 60, pp. 211–221
Chai, T., and Draxler, R.R. 2014. Root mean square error (RMSE) or mean absolute error (MAE)? – Arguments against avoiding RMSE in the literature. Geoscientific Model Development, Vol. 7, pp. 1247–1250.
de Myttenaere, A., Golden, B., Le Grand, B. and Rossi, F. 2015. Mean Absolute Percentage Error for Regression Models. Neurocomputing, Vol. 192, pp. 38–48
Drewes, H, 2010. Eye Gaze Tracking for Human Computer Interaction. PhD dissertation der Ludwig-Maximilians-Universität München.
Duchowski, A., 2007. Eye Tracking Methodology – Theory and Practice. London: Springer-Verlag.
Kenyon, R.V., 1985. A Soft Contact Lens Search Coil for Measuring Eye Movements. Vision Research, Vol. 25, No. 11 pp. 1629-1633.
Schall, A. and Bergstrom, J.R. 2014. Eye Tracking in User Experience Design. Amsterdam: Elsevier.
Sharma, B. and Venugopalan, K. 2014. Comparison of Neural Network Training Functions for Hematoma Classification in Brain CT Images. IOSR Journal of Computer Engineering, Vol. 16, No. 1, pp. 31–35.
Simola, J., Saloj?rvi, J. and Kojo, I, 2008. Using Hidden Markov Model to Uncover Processing States from Eye Movements in Information Search Tasks. Cognitive Systems Research Vol. 9, pp. 237–251.
Zhong, M., Xinbo, Z., Xiao-chun, Z., Wang, J. Z. and Wenhu, W. 2014. Markov Chain Based Computational Visual Attention Model that Learns from Eye Tracking Data. Pattern Recognition Letters, Vol. 49, pp. 1–10