Analyzing the performance of fuzzy cognitive maps with non-linear hebbian learning algorithm in predicting autistic disorder

https://doi.org/10.1016/j.eswa.2010.06.069Get rights and content

Abstract

The soft computing technique of fuzzy cognitive maps (FCM) for modeling and predicting autistic spectrum disorder has been proposed. The FCM models the behavior of a complex system and is used to develop new knowledge based system applications. FCM combines the robust properties of fuzzy logic and neural networks. To overwhelm the limitations and to improve the efficiency of FCM, a good learning method of unsupervised training could be applied. A decision system based on human knowledge and experience with a FCM trained using unsupervised non-linear hebbian learning algorithm is proposed here. Through this work the hebbian algorithm on non-linear units is used for training FCMs for the autistic disorder prediction problem. The investigated approach serves as a guide in determining the prognosis and in planning the appropriate therapies to special children.

Introduction

Autism is a brain development disorder that first gives signs during infancy (or) childhood and generally follows a steady course without remission or relapse. Autism is characterized by widespread abnormalities of social interaction and communication and severely restricted interests and highly repetitive behavior (American Psychiatric Association, 2004). The main characteristics are impairments in social interaction, impairments in communication, restricted interests and repetitive behavior (Cohen & Donnellan, 1987). They also qualitatively differ in their use and understanding of non-verbal activities that influence the interaction with others such as eye contact, posture, facial expression and gestures. Persons with autism have a restricted range of interests, where as their thinking and behavior is rigid. In a recent study, a discriminant function analysis of the magnetic resonance imaging (MRI) of brain measures were used for the classification of autistic spectrum disorders (Akshoomoff et al., 2004). Numerous algorithms and methods have been proposed in order to achieve data classification and to improve its efficiency (Ishibuchi et al., 1993, Papageorgiou et al., 2008a, Sabeti et al., 2007). A recent study proposes a new algorithm in predicting autistic disorder using neuro fuzzy model and hybrid model (Arthi and Tamilarasi, 2008, Arthi and Tamilarasi, 2009a, Arthi and Tamilarasi, 2009b). But the fuzzy cognitive map approach is a different technique from neural networks due to their properties and knowledge elicitation from experts (Papageorgiou et al., 2003b, Papageorgiou et al., 2007). It is an alternative approach to knowledge based techniques offering more semi-quantitative characteristics. The decision making problem of predicting autistic disorder is a complex process, because of the numerous elements/parameters (such as symptoms, signs-movements, etc.) involved in its operation, and a permanent attention is demanded. The knowledge of physicians according to the children’s signs and symptoms is the main point to succeed a diagnosis. In this study, the fuzzy cognitive mapping (FCM) approach is investigated to handle with the problem of autistic disorder prediction. The main topic of the presented effort is the representation of the cause-effect relationships within medical data by the application of the soft computing technique of fuzzy cognitive maps and the training of these relationships using an efficient unsupervised learning algorithm. FCM is a knowledge-based approach exploiting the main features of fuzzy logic and neural networks, and thus corresponding to an artificial cognitive network. It can handle efficiently with complex modeling problems to assess medical decision making tasks.

In this proposed work, a new implementation approach on training fuzzy cognitive maps with non-linear hebbian learning is investigated and a clinical problem of ASD with real cases is examined to analyze the process and justify the results. Expert’s involvement is essential and obvious in this proposed model. Experts are the good assessors of this neurological disorder to promote the better performance of this framework. They can handle with the available experience and accumulated knowledge from experts. The easy of use and the low time requirement are important features of FCMs. Here, the main aim of this paper is to create a model that could predict the category of autism using the unsupervised non-linear hebbian learning algorithm for FCM tool. The overall structure of the decision making approach based on FCMs is investigated to help physicians, through the design of the knowledge representation and reasoning with FCM for the problem of autistic disorder prediction. Further, a number of different scenarios concentrated on autistic disorder individually for real clinical cases are examined to demonstrate the application of the proposed methodology and its functioning. The paper is organized into the following sections: the second section provides the main aspects to the fuzzy cognitive map theory and presents a description on the FCMs for medical decision making tasks. This is followed by the description of the problem of ASD, the reasons why the FCM approach was considered appropriate for modeling this particular domain and the stages in the development of the FCM model, constructing the resulting model for ASD. The fourth section provides a description on non-linear hebbian learning algorithm for training FCM for the specific problem of ASD. The next section provides its simulation analysis through a number of different scenarios and discussion of results. The conclusion is followed by exploring the potential of the FCM as a dynamic model in medicine for making decisions.

Section snippets

Fuzzy cognitive maps in decision making

Fuzzy cognitive maps are diagrams used as causal representations between knowledge/data to represent events relations. They are modeling methods based on knowledge and experience for describing particular domains using concepts (variables, states, inputs, outputs) and the relationships between them (Kosko, 1986). FCM can describe any system using a model having signed causality (that indicates positive or negative relationship), strengths of the causal relationships (that take fuzzy values),

Autistic disorder problem description and construction of fuzzy cognitive map model

Here, in this subsection, the ASD problem is described consisting of the 24 concepts presented in Table 1. Three experts like pediatrician, occupational therapist and special educator were used to determine the main concepts of the model as well as their interconnections among concepts. This process was completed through a questionnaire, which created for this process and proposed to the team of experts. The questionnaire is presented in Appendix A which is from the modified checklist for

Non-linear hebbian learning algorithm

The Hebbian paradigm is perhaps the best-known unsupervised learning theory in connectionism (Papageorgiou et al., 2006). The non-linear hebbian learning algorithm focus in the domain of artificial neural network field and it embodies properties such as locality and the capability of being applicable to the basic weight-and-sum structure of neuron models. Papageorgiou et al., proposed a new method for characterizing brain tumors using unsupervised hebbian learning algorithm (Papageorgiou et

Simulation results and discussion

All the 23 concepts are considered as factor concepts by the experts as they are listed in Table 1 to design FCM model and each concept represents three fuzzy values. In this problem, the concept C24 have been considered from the experts as decision output concept – DOC and could be categorized as Definite Autism (DA), Probable Autism (PA) and No Autism (NA) which takes a range of values such as 0.41  DA  1.00, 0.26  PA  0.40 and 0  NA  0.25, respectively. For this FCM model of the ASD problem, the

Conclusion

The knowledge-based approach used in this work focuses on the soft computing technique of fuzzy cognitive maps with NHL training algorithm for the estimation of medical outcomes and resource utilization. In the presented research, we have proposed the FCM learning approach for the process of predicting the autistic disorder. The presented solution has been raised by some of the requirements imposed by the targeted application: the causal association of symptoms and signs of autistic spectrum

Acknowledgements

Authors would like to thank the Dr. A. Thejavathi, pediatrician and Ms. Sudha, Occupational therapist, Coimbatore for their kind support in collection of datasets.

References (28)

  • D.J. Cohen et al.

    Handbook of autism and pervasive developmental disorders

    (1987)
  • Georgopoulos, V. C., & Stylios, C. D. (2003a). Augmented fuzzy cognitive maps based on case based reasoning for...
  • H. Ishibuchi et al.

    Neural network that learn from fuzzy If-then rules

    IEEE Transactions on Fuzzy Systems

    (1993)
  • Y. Miao et al.

    On causal inference in fuzzy cognitive maps

    IEEE Transactions on Fuzzy Systems

    (2000)
  • Cited by (91)

    • Learning large-scale fuzzy cognitive maps using an evolutionary many-task algorithm

      2021, Applied Soft Computing
      Citation Excerpt :

      Most of existing methods were proposed to overcome small-scale FCM learning problems [24]. The Hebbian-based algorithms could learn a reliable FCM by fine-tuning the causal connection specified by the expert, including nonlinear Hebbian learning (NHL) algorithm [25], active Hebbian learning (AHL) algorithm [26], data driven nonlinear Hebbian learning (DDNHL) algorithm [27], and so on. The population-based algorithms, such as real-coded genetic algorithm (RCGA) [28], ant colony optimization (ACO) [29], particle swarm optimization (PSO) [30], and competent memetic algorithm (CMA) [31], have attracted many attentions among scholars.

    • Fuzzy logic-based hybrid knowledge systems for the detection and diagnosis of childhood autism

      2021, Handbook of Decision Support Systems for Neurological Disorders
    View all citing articles on Scopus
    View full text