Designing a Model for Predicting and Evaluating the Innovation Capacity of Knowledge-based Companies with a Neural-Adaptive Fuzzy Inference System (ANFIS).

Assessing the innovation capacity of knowledge-based companies and predicting their innovation capacity is very important for these companies, and the decision to transfer or expand the company's technology depends on the level of innovation capacity. The main purpose of this study is to design a model for assessing the innovation capacity of knowledge-based companies with a neural-adaptive fuzzy inference approach. Nervous-adaptive fuzzy inference system (ANFIS) is a good way to solve nonlinear problems. ANFIS is a combination of fuzzy inference and neural network that utilizes both. The research and statistical sample population for compiling, implementing and testing the model is all the knowledge-based companies of Pardis Technology Park, and finally 180 items were evaluated, collected by expert evaluators and based on model calculations. To evaluate the performance of the model, the parameters of the average error square (RMSE), relative error percentage (e), absolute error average (MAE) and coefficient of explanation (R2) were calculated, which are 0.0136, 1.3%, and 0.048, respectively. And 0.998 was obtained. which indicates the accuracy and reliability of the model output prediction. This research is descriptive-survey in terms of purpose, application, and data collection method. The output of this study is "ANFIS".

Assessing the innovation capacity of companies is a complex and elusive concept that is difficult to determine. Measuring the capacity for innovation requires quantitative and qualitative considerations. In both cases, the companies are both. And .. Innovative. Therefore, knowledge-based companies that are leading the new economy in the world need to recognize their innovation capacities and be aware of the level of innovation capacity and formulate specific strategies to reach the desired level. The knowledge economy has been the mainstay of investment in small and medium-sized enterprises. In Europe, these companies make up 99.8% of the total number of companies in EU member states, numbering 19 million across the EU. (E-Business policy group 2016). Knowledge-based companies that are constantly moving on the edge of knowledge, their products and services are innovative. These companies have the right innovation capacity to make the necessary decisions about the company's technology transfer or development by relying on that capacity. What managers are interested in is the level of innovation capacity of the company and predicting the level of innovation capacity according to the trend of changes in innovation capacity indicators.

The statistical population of this study in the process of identifying and extracting the components and dimensions of innovation capacity, includes 19 professors in the field of innovation and experts in knowledge-based companies. The statistical sample for compiling and implementing the model is 241 knowledge-based companies located in Pardis Technology Park.

The innovation of this research is in the method of research, because according to some features of the concepts of innovation and capacity measurement, both of which have their own ambiguities and ambiguities, their study was not accurate through binary science (Azar and Faraji, 2017). Therefore, by examining the existing literature (Wang's method) and using the opinions of experts, the researcher has chosen the approach of fuzzy systems that can model the qualitative aspects of human knowledge and reasoning processes without using a little precise analysis using one of the fuzzy rules. Famous fuzzy inference, for modeling, fuzzy-nerve inference method Or the ANFIS is comparative.

This research is based on the purpose, applied and developmental type (using the model to measure the amount and predicting the innovation capacity in the company) and according to the method of data collection and complete step-by-step research of descriptive-survey type (which is a method for obtaining information). It can also be explored in terms of the views, beliefs, opinions, behaviors, motivations, or characteristics of a group of members of a community (as well as the discovery of the dimensions and components of innovation capacity). The reason for using ANFIS in both systems is as follows: (Ata & Kocyigit, 2010) 1-Use neural networks to sort data and identify patterns. 2-Creating a clear fuzzy inference system that has fewer problems and fewer errors in computations than neural networks. 3-This system maintains the benefits of a fuzzy expert system, while also reducing the need for an expert. 4-Due to the use of fuzzy logic, the problems of modeling and analyzing complex data in this method are reduced. 5-It is possible to enter the qualitative dimensions of human experience into this system. 6-Finally, the neural fuzzy system has the ability to learn while maintaining the benefits of the fuzzy inference system. According to the conceptual model, the following can be developed and designed in two ways: neural-adaptive fuzzy inference model: 1) The first way is the direct use of 25 indicators as model inputs (the advantage of this method is the awareness and direct observation of the role of each indicator in the model, but the complexity and length of calculations are its disadvantages). 2) The second way is to use the indicators in the form of 5 dimensions to form smaller and simpler models in order to use the output of these subsets as inputs to the final model (the advantage of this way is the simplicity of the model but no direct knowledge of the role of each indicator in the model). In his research, the researcher has calculated the innovation capacity in both ways. In this paper, the first method that is based on indicators is presented. For this purpose, the data of 180 companies were evaluated by experts based on the above 25 indicators, a 180 * 26 matrix (26 columns and 180 rows, the last column as the score of each company) was prepared. By defining a random data selection function, the data matrix was divided into two separate parts with optional and completely random percentages, 80% of the data was used for system training and 20% for model testing. This fuzzy inference system is able to provide analyzable outputs based on the inputs provided. Analytical output helps the company to assess the sensitivity of the output by changing each of the input variables and make the necessary corrections based on it.

 The validation of the research was considered from two perspectives; first, validation of data collection tools, which was confirmed by the opinion of narrative experts and SPSS software reliability of the input indicators. Expert evaluation has been used. To evaluate the performance of the model, the parameters of the mean error square error (RMSE), relative error percentage (e), absolute error average (MAE) and explanatory coefficient (R2) were used and the values of 0.0133, 1.3%, 0.048 and 0.998 were obtained, indicating the accuracy and reliability of the model output prediction.

Some aspects of this research are similar to the dimensions of Morel and Bowley's research (2004-2014), such as research and development activities, ideas, creativity and customer relationship, but the indicators of these dimensions are completely different according to the statistical community and technological differences between the two countries. They are different. Also, compared to Wang's research, although the method used is fuzzy, the method used in this research is the fuzzy neuro-adaptive inference method, which has the greatest advantage for model design. Compared to Proshch et al.'s (2017) research, indicators such as total scientific and technical staff (S&T), total R&D spending, and risk capital (VC) are common. In Hayata Research (2018), employee quality indicators, research capacity and multiplicity, budget and research grants, and access to the research community are shared. Compared to the research of Arasti et al., There are only common dimensions in the field of dimensions, but the indicators of these dimensions are completely different. During the research process, many ideas and opinions came to the researcher's mind that there was no opportunity to implement and implement them. Therefore, the following topics are recommended for interested researchers in this field: •Connect the fuzzy inference system to the organization's management database to receive instantaneous information and provide analytical reports on the status of innovation capacity in the form of intelligent business systems. •Use simulation algorithms and evolutionary algorithms to analyze and simulate implementation steps and provide effective reports to managers. •Combining neural networks and genetic algorithms to design the optimal model for implementing an innovation capacity assessment system. •Combining fuzzy logic, neural network, genetic algorithms, and meta-innovative optimization methods to produce an intelligent system that is continuously able to provide improving suggestions for improving the evaluation system.