Elsevier

Neurocomputing

Volume 74, Issue 17, October 2011, Pages 2834-2841
Neurocomputing

Questionnaires-based skin attribute prediction using Elman neural network

https://doi.org/10.1016/j.neucom.2011.03.040Get rights and content

Abstract

Skin attribute tests, especially for women, have become critical in the development of daily cosmetics in recent years. However, clinical skin attribute testing is often costly and time consuming. In this paper, a novel prediction approach based on questionnaires using recurrent neural network models is proposed for participants’ skin attribute prediction. The prediction engine, which is the most important part of this novel approach, is composed of three prediction models. Each of these models is a neural network allocated to predict different skin attributes: Tone, Spots, and Hydration. We also provide a detailed analysis and solution about the preprocessing of data, the selection of key features, and the evaluation of results. Our prediction system is much faster and more cost effective than traditional clinical skin attribute tests. The system performs very well, and the prediction results show good precision, especially for Tone.

Introduction

Skin attribute testing is an important task in daily cosmetics development. The traditional clinical skin attribute test is carried out by clinical experiments, which not only require large amounts of information, complicated processes, but also need expensive specialized equipment. Therefore, a rapid and low cost test method is needed for skin attribute evaluation in cosmetics stores and other places. Fortunately, information technology has given hope to cosmetics company research and development (R&D) to predict skin attributes based on existing experimental data in a new approach.

Boelsma et al. [1] have used questionnaires to evaluate whether or not those features they are interested in are associated with the skin attribute of humans. Grove et al. [2] have also used questionnaires to evaluate the products' effects on the skin attributes in the R&D process. However, to the best of our knowledge, no one has realized questionnaire-based skin attribute prediction with a data mining method. Most of the current work achieves skin attribute prediction through the use of questionnaires without data mining methods. They usually select questions according to their own experiences. In the prediction process, simple logic is used to judge a portion of answers. Those prediction methods are linear predictors, while skin attribute is a nonlinear function of its input features, and the results in the existing work are usually inaccurate.

In this study, a novel prediction approach based on recurrent neural network models is proposed for participants’ skin attribute prediction. Three realization steps must be completed to achieve skin attribute prediction with data mining methods in our prediction system. The first is digitizing the existing experimental data and questionnaires, and transforming them into numeric or several nominal data that can be processed by the computer. Then the key features of skin attributes are selected with the digitalized data. With the key features, the prediction models for skin attributes are constructed. Women's skin attributes can be predicted by computer directly with these models. In response to these problems, we give a total solution by using the data preprocessing, feature selection and prediction methods in the area of data mining.

This new prediction approach has three main characteristics. The first one is that it saves time. The only thing participants must do is fill out the questionnaire instead of taking complex clinical tests. The second is that it saves cost. The results are given by computer immediately without complex manual analysis. The third is that this prediction approach is very simple. Participants can complete the questionnaire at their home via the internet. This new approach allows participants to understand their skin attributes much easier. Skin attribute research work will also become easier.

The remaining parts of this paper are organized as follows: Section 2 covers some related work including data preprocessing methods, feature selection methods and the neural network model. The design of the proposed prediction model is presented in Section 3. The system implementation is shown in Section 4. In Section 5, the experimental results are shown and evaluated. Finally, Section 6 presents our conclusions.

Section snippets

Related work

In this section, we survey some related research techniques in the area of data preprocessing, feature selection and prediction. Before skin attribute prediction is implemented with the existing questionnaires and clinical experimental data, data must be preprocessed. Information needs to be extracted from the original questionnaires so that it can be handled by computer. A brief analysis of distribution of the original data also needs to be made. After preprocessing, key features are selected

Data preprocessing

The raw questionnaire data must be transformed into the form that can be handled by computer, so data preprocessing is very important before key features selection. The work in this paper is based on two questionnaires. One is focused on the testers' basic information, such as age, educational background, life habits and occupations. The other investigates the testers' habits of using sunscreen and how much she knows about UV products. The testers' skin attributes (Tone, Spots and Hydration)

System implementation

Based on the preceding skin attribute prediction method, a skin attribute prediction system for Chinese women has been developed. In the preparation phase of the system, these questionnaires about the Chinese women's skin attributes are preprocessed first. Then key features are selected from the processed data. According to different skin attributes' key features, the corresponding data is normalized to obtain the training data set, with which different skin attributes’ prediction models are

Key features

In this paper, the questionnaires are filled out by Chinese women living in the urban area of Beijing and Guangzhou. The original questionnaires have a total of 49 questions and contain 905 instances. These data are used to select key features of three different targets—Tone, Spots, and Hydration. In the original questionnaires, one question asks whether the tester used facial moisturizer today or not. Because facial moisturizer may make obvious changes to the tester's hydration, we removed the

Conclusions

In this paper, we propose a novel prediction approach based on neural network model that helps predict the skin attributes for participants. This new prediction approach contributes a great deal to skin attribute testing, which reduces much of the time and costs needed to conduct experiments. If a participant wants to test her skin attributes, the only thing she needs to do is complete the questionnaire instead of undergoing traditional complex clinical tests. This prediction approach is so

Acknowledgments

This work is supported by P&G–Tsinghua Collaboration Research Funding, National Natural Science Foundation of China (Grant No. 60875073), and Important National Science & Technology Specific Projects (Grant No. 2009ZX02001). We would like to thank Amy Chastain for reviewing and polishing the language of this paper. Thanks also to all reviewers for their constructive comments and guidance in shaping this paper.

Wei Wan is currently pursuing his master degree in the State Key Laboratory on Intelligent Technology and Systems, Department of Computer Science and Technology at Tsinghua University in Beijing, China. He also graduated from the Department of Computer Science and Technology at Tsinghua University in 2009. His research interests span the areas of parallel processing, data mining, machine learning, and natural language processing.

References (27)

  • J.L. Elman

    Finding structure in time

    Cognitive Science

    (1990)
  • R.S. Toqeer et al.

    Speed estimation of an induction motor using Elman neural network

    Neurocomputing

    (2003)
  • C.-Y. Liou et al.

    Modeling word perception using the Elman network

    Neurocomputing

    (2008)
  • Cited by (9)

    • Computational face reader based on facial attribute estimation

      2017, Neurocomputing
      Citation Excerpt :

      In [28], the authors follows this idea to simultaneously learn the location information and the values of facial attributes. Besides on the above works for facial attribute estimation, there are some works focus on estimating the attributes of body appearance, such as skin color [29], cloth color, types of cloth (e.g., jackets, skirts, trousers, etc.) [30], standing position, and so on. To the best of our knowledge, this work is the first attempt to study the face-reading problem in computer vision field.

    • Adaptive backstepping Elman-based neural control for unknown nonlinear systems

      2014, Neurocomputing
      Citation Excerpt :

      Because of the context neurons and local recurrent connections between the context layer and the hidden layer, the Elman neural network has certain dynamic advantages over a static neural network. Although a typical Elman neural network has found various applications in speech recognition and time series prediction [15,16], its training and convergence speed are usually very slow and not suitable for time critical applications. To solve this problem, several modified Elman neural networks have been proposed for the identification and control of an unknown nonlinear system [17–19].

    View all citing articles on Scopus

    Wei Wan is currently pursuing his master degree in the State Key Laboratory on Intelligent Technology and Systems, Department of Computer Science and Technology at Tsinghua University in Beijing, China. He also graduated from the Department of Computer Science and Technology at Tsinghua University in 2009. His research interests span the areas of parallel processing, data mining, machine learning, and natural language processing.

    Prof. & Dr. Hua Xu received his B.S. from Xi'an Jiaotong University in 1998. He received his M.S. and Ph.D. from Tsinghua University in 2000 and 2003, respectively. Now he is working in dept. of C.S., Tsinghua University. His research fields include the following aspects: data mining, intelligent information processing and advanced process controllers for IC manufacturing equipments. He has published over 30 academic papers, received 3 invention patents of advanced controller and is also the copyright owner of 5 software systems.

    Wenhao Zhang is a Ph.D. student in Department of Computer Science and Technology in Tsinghua University in Beijing, China. Previously he received his B.E. in Software Engineering from Software School of Dalian University of Technology in 2010. His current research interests are on data mining, and semantic web, natural language processing, in particular in sentiment analysis.

    Dr. Xincheng Hu received his Ph.D. in Environmental Chemistry, from Osaka Prefecture University, Japan. Now he is working in Skin Care in P&G Technology (Beijing) Co., Ltd. engaged in the research and development for skin care product formulation, clinical study for 16 years.

    Dr. Deng Gang received his B.S., Master and Ph.D from Nankai University. Joined Procter & Gamble company in 2006, senior scientist in clinical function. His research fields include the following aspects: Asian female and male skin understanding, whitening active screening, skin modeling, and skin measuring technology development. He has published over 15 academic papers in biology area.

    View full text