Predicting the Intention to Adopt Innovation in Supply Chain Finance: Determinants of Brazilian FinTech

Based on the mixed model unified technology acceptance and utilization theory (UTAUT) and spinner innovation model (SPINNER), a theoretical model is suggested to explain the determinant of behavioral intention to predict innovation in the context of a financial sector firm. A questionnaire was developed to collect primary data, which was subsequently processed through the artificial intelligence technique (deep learning). The constructs (performance expectancy, effort expectancy, social influence, facilitating conditions, behavioral intention, public knowledge, private knowledge, and innovation) supported the model, including mediating hypotheses. It was observed that the mixed methodological approach (SEM and ANN) can help to find the linear and non-linear relationships better, being that the error of the predicted model is 0.104, that is, 10.4% relatively low, which evidences that ANN can be used to predict the dependent variable innovation safely.


INTRODUCTION
Supply chain financial services have been growing in literature.For instance, a robust Supply Chain Financial Logistics Supervision System was constructed by harnessing the power of Internet of Things (IoT) technology to examine the practical implementation of IoT technology in enabling customers to supervise the logistics process effectively (Liu et al., 2023).Also, a comprehensive supply chain finance (SCF) framework introducing two novel coordinating contracts that leverage trade credit financing was designed to address different problem settings within the supply chain (Emtehani et al., 2023).
FinTech has gained popularity as it refers to innovative technologies adopted by financial service institutions.The emergence of online peer-to-peer (P2P) lending platforms has introduced a promising FinTech business model that connects investors with capital recipients within supply chains (SCs) (Taleizadeh et al., 2022).Recent advancements in financial technology, known as FinTech, have emerged as solutions to various challenges.These FinTech-driven business models, including crowdfunding, peer-to-peer lending, invoice trading, mobile wallets and payments, and platformdriven SCF, are reshaping the landscape for small businesses (Chang et al., 2021;Chen, Li et al., 2021;Leung, Cho, & Wu, 2022;Liu Panfilova et al., 2022;Liu, Sakulyeva et al., 2022;Malaquias et al., 2021;Shankar, 2022;Sharma et al., 2023;Wamba et al., 2021).The effects of dynamic employee capabilities, FinTech, and innovative work behavior on employee and supply chain performance in the Vietnamese financial industry were analyzed in terms of impact (Phan et al., 2022).
The "prediction" and "adoption" approaches have been gaining cult in the FinTech market.Using structural modeling equations and neural networks has become popular in the financial sector by researchers aiming to identify users' behavior intentions of digital services.The "single" methods have a fundamental difference in relation to the mixed approach, as they are based on a single approach and depend on a single model built based on the acquired knowledge.For example, partial least squares structural equation modeling was used to understand the intention of behavioral use of the FinTech services by companies, a causal-predictive analysis (Irimia-Diéguez et al., 2023).A correlation-regression analysis scenario method for forecasting was used to describe the number of FinTech companies in the finance sector (Taujanskaitė & Kuizinaitė, 2022).In addition, a deep learning-based prediction model is implemented to predict the price movement of fund classes based on the classification results in China FinTech (Chen et al., 2021).
The unified technology acceptance and utilization theory (UTAUT) model has been used to investigate behavioral intentions in several contexts.Recently, studies examined the behavioral intentions of extension professionals from two extension systems to foster the adoption of precision farming (Lee et al., 2023).Prior studies have examined the factors that impact the acceptance of mobile learning technology for 21st-century skills-based training among teachers in Saudi Arabia and Pakistan (Dahri et al., 2023).
In addition, Sulistyaningrum et al. (2023) focus on integrating three theories, including the UTAUT model, to determine factors that could influence people's behavior toward adopting telepharmacy services.Also, another application of the model was developed to investigate the main factors influencing users' intention to accept e-wallets in Jordan (Hammouri et al., 2023).Conversely, the Spinner Innovation Model (SPINNER) has been used to predict small and medium enterprises (SMEs) innovations.Prior studies explore the prediction of innovation in SMEs applying the data mining technique -cross-industry standard process for data mining (CRISP-DM) (Figueiredo et al., 2023).Recently, Figueiredo et al. (2023) explored the integration between SPINNER and the Triple Helix Model to analyze the influence to improve system innovation in SMEs.Equally important, Figueiredo and Ferreira (2020) applied SPINNER to explore Brazil's innovation propensity in the service sector.The model integrated the variable knowledge creation, knowledge transfer, and innovation.Figueiredo et al. (2020) explored the innovation and co-creation in KIBS with the Spinner Innovation Model application.They found that SPINNER explained the propensity for innovation in KIBS and demonstrated that the innovation process was based on the knowledge integrated with co-creation and knowledge transfers.The study contributes to the literature by addressing a new model to predict the intention to adopt innovation in SCF in the Brazilian financial sector, specifically FinTech.However, previous studies may show non-representative results because they were applied in isolated ways in different economic sectors and countries.For example, Zhong and Hitchcock (2021) sought a new predictive model for stock price forecasting to solve the problem of insufficient data for machine learning models.In addition, Wang et al. (2022) addressed the credit risk prediction in small firms in SCF, showing that financial-based information (e.g., TOC and NIR) are more useful in predicting the credit risk of SMEs in SCF.Erfanian et al. (2022) investigated macroeconomic and microeconomic indicators in the SCF and defined that the BTC price prediction is feasible by means of economic theories.
To address the research gaps, our study explores the following question: How do we predict the intention to adopt innovation in SCF in the Brazilian FinTech?
The motivation for the study to explain the determinant of behavioral intention to predict innovation in the context of a financial sector firm is based on the following factors: i) the financial sector, especially FinTech, has grown in several markets, including Latin America; thus, it is interesting to investigate the application of a predictive mixed model, including two analysis techniques, since they are considered of high capacity of exploration and analysis of data applied in the scientific literature; ii) when compared to traditional data analysis techniques, the application of mixed models in predicting the intention to innovate in the financial sector is still in its early stages.
Even though many researchers have addressed the two models in several different studies, the accuracy of the mixed model presented in our study can help to find the linear and non-linear relationships better, being that the error of the predicted model is 0.104, that is, 10.4% relatively low, which evidences that Artificial Neural Network (ANN), can be used to predict the dependent variable innovation safely; iii) when it comes to predicting the intention to innovate in finance, the proposed model becomes an important tool.
The prediction of the adoption of innovation in the financial sector, FinTech, is extremely relevant because it collaborates with the premise of the market, innovation as a factor of competitiveness and exponential growth (Talay et al., 2017).In addition, the study addresses the contribution of the variables of the UTAUT and SPINNER models, considering the prediction of innovation adoption in the Brazilian market, which grows exponentially.Our approach developed for the model presents significant accuracy in predicting innovation considering future scenarios, thus being unprecedented in the international literature.
The presented study is organized as follows.In the next section, we present a literature review on innovation and SCF, taking an initial approach, reviewing previous studies, and summarizing the empirical results from the adopted models.Then, we discuss how the conceptual modeling was developed from two scientific models, the UTAUT and the SPINNER models, applied respectively to technology adoption and innovation forecasting and includes the variables used in the ANN model.Next, we show the results based on the empirical model and discuss the findings with the previous literature.Finally, we present the main conclusions, limitations, and future research.

LITERATURE REVIEw AND HyPOTHESES DEVELOPMENT Predict Innovation and Supply Chain Financing in FinTech Firms
FinTech firms serve a pivotal function as intermediaries in signaling, enabling interactions between pertinent stakeholders, expediting the exchange of information, and mitigating information disparities stemming from data overload (Song et al., 2023).Furthermore, the advancement of financial technology (FinTech) has bolstered the embrace of SCF by utilizing information technologies (IT) to provide financial services and streamline lending and transaction procedures for SMEs (Soni et al., 2022).
Furthermore, previous studies have applied and developed various traditional prediction methods for innovation in the financial sector.For example, Figueiredo et al. (2023) investigate the potential impact of integrating Spinner Innovation and Triple Helix models on enhancing systemic innovation.Li et al. (2022) explained the correlation between operational capabilities and investment components to aid firm managers and investors in comprehending the innovation process in SCF.
Among them, many are applied to FinTech to determine the main factors determining the change in FinTech's behavior, considering the economic sector's prediction and future development in digital services (Taujanskaitė & Kuizinaitė, 2022).For instance, Abou-Shouk and Soliman (2021) used the Unified Theory of Acceptance and Use of Technology model to explore the factors that lead to and result from the intention of tourism organizations to adopt gamification in customer engagement.
An earlier study developed a forecasting model based on machine learning to predict credit risk in SMEs in China using financial information, operation information, innovation information, and negative events as predictors (Wang et al., 2022).In addition, Horvát et al. (2023) found that Heckman's two-variable model (speed and diversity of opinion) predicts who is funded and who repays the financial outcome.Using correlation analysis, Li and Tan (2021) described that the ARIMA model predicts the actual value using the "Operating income growth rate" indicator.
The Taskforce on Nature-related Financial Disclosures (TNFD Framework) is another traditional approach to predicting better risk management from health firms.Deweerdt et al. (2022) discovered that many factors can stifle innovation because companies cannot predict the return, and innovation is complex and unclear to achieve.Moreover, Erfanian et al. (2022) included ordinary least squares (OLS) and multilayer perceptron (MLP) to predict the BTC price.Wu (2023) explored the influence of big data technology on the innovation enterprise economic mode, showing that the company's planned deployment of staff deviates only slightly from the actual deployment of staff.
In addition, Zhong and Hitchcock (2021) used the variables, weekly historical prices, finance reports, and text information to apply a predictable model for stock prices.Also, Liu et al. (2021) conclude that blockchain crowdfunding, FinTech, encryption currency, and SCF are the key research directions in the study.Regarding how stock market reactions to marketing actions affect subsequent marketing decisions, Talay et al. (2017) describe that the available literature on marketing finance, especially on innovation, affects the predictive power of immediate actions.However, the literature review shows the importance of FinTech´s studies to financial sector products, including innovative methods to engage clients (Garg et al., 2023).Considering SCF to support baking systems, Mahmoudi et al. (2023) applied a multiple-criteria decision-making (MCDM) problem and proposed a model based on the Ordinal Priority Approach (OPA).
Looking at the E-commerce SCF for SMEs in the role of green innovation, Guo et al. ( 2023) used regression analysis techniques to have the results.In addition, numerous financial institutions are keen on embracing technological solutions to bolster operational efficiency in managing SCF.This intricate process involves multiple participants and encompasses many complex financial activities (Kao et al., 2022).Regarding future research, Sharma et al. (2023) considered that FinTech addresses some challenges regarding platform-driven SCF in the ecosystem context.
However, traditional models have been used to describe some results in the context of the SCF Business Model, bringing mixed theoretical analysis of the Business Model Canvas (BMC) to the perspective of competitive advantages (Zhou & Lee, 2023).In contrast, advanced technology (blockchain) has been used to address current challenges in supply chain financing processes (Tsai, 2023).This means that academics and professionals urgently need accurate and up-to-date knowledge of FinTech (Phan et al., 2022).
One more method is decision-making used for different scales in the ARDL long-term coefficients and AHP to financial investment decisions (Atmaca & Karadaş, 2020), and the accurate forecasts for investment decision-making could be used in terms of market returns as the most effective tools for risk management (Mallikarjuna & Rao, 2019).
To investigate how financial literacy and behavioral traits affect the adoption of electronic payment (ePayment) services, Long et al. (2023) used the instrumental variable approach.In addition, (VARFIMA) model was used by Oral and Unal (2019) to model and extract the time series to estimate the forecasting process.Also, the grey system theory was used to predict the Bitcoin price changes (Faghih Mohammadi Jalali & Heidari, 2020).In addition, Big data analytic techniques associated with machine learning algorithms were an important application in the stock market investment field (X.Zhong & Enke, 2019).
Finally, the Group Method of Data Handling (GMDH) neural network has exhibited strong performance in data mining, prediction, and optimization (Zhang et al., 2023).

HyPOTHESES Performance Expectancy
Performance expectancy can be defined as a belief related to the ability to succeed in an action (Wu & Kang, 2021).Performance expectancy refers to how much technology aids consumers in accomplishing specific tasks (Fedorko et al., 2021).Also, the level of performance expectancy associated with open data (OD) plays a crucial role in the user technology acceptance models, especially concerning the future implementation of OD in Industry 4.0 and its potential impact on Society 5.0.(Sołtysik-Piorunkiewicz& Zdonek, 2021).In addition, performance expectancy can be described as the anticipated influence of a technology's functional advantage, even when operating in uncertain conditions (Sewandono et al., 2023).
Hypothesis 1. Performance Expectancy (PEXP) positively influences the intention to predict innovation.

Effort Expectancy
Performance expectancy represents how much technology assists consumers in carrying out specific actions.On the other hand, effort expectancy pertains to the ease with which consumers can utilize the technology (Fedorko et al., 2021).In addition, The relationship between digital competence and effort expectancy concerning work engagement has not been adequately explored or understood (Sang et al., 2023).Effort expectancy refers to the level of ease experienced by individuals while using technology (Venkatesh, 2022).Moreover, studies have revealed a significant impact of effort expectancy on the elderly's intention to use technology (Ramírez-Correa et al., 2023).
Hypothesis 2. Effort Expectancy (EEXP) positively impacts the intention to predict innovation for supply chain financing.

Social Influence
Social influence is a means of getting more useful information about the target product (Yang et al., 2023).In the medical field, social influence can be considered one-factor affecting therapy retention (Knight et al., 2023).According to Zareie and Sakellariou (2023), social influence is formed through human interactions and can significantly impact shaping opinions, facilitating the rapid and extensive dissemination of specific messages or news, and expediting the formation of collective viewpoints.Also, social influence plays a crucial role in developing call comprehension abilities (Garcia-Nisa et al., 2023).
Hypothesis 3. Social Influence (SIN) positively explains the intention to predict innovation for supply chain financing.

Knowledge
Private knowledge can extract local knowledge from data distributed in a decentralized way to collectively develop an intelligent model with differentiated privacy guarantees (Qi et al., 2023).In this configuration, each client retains its training data locally, without sharing it externally, while a central server maintains the intelligent model and provides a local copy to each client.However, public and private knowledge in product development has introduced new collaboration arrangements in which existing resources and knowledge combine with those traditionally retained by local and global companies (Ferpozzi, 2023).In addition, at certain times, the private sector, with limited knowledge, leads the search for new partners to develop available services based on confidentiality and a desire for personalized service (V et al., 2022).Finally, numerous studies in knowledge management suggest that enterprises and organizations can reap significant benefits by establishing systematic sharing, transfer, and reuse of knowledge.While knowledge sharing and transfer require effective mechanisms for capturing and transferring information, little research has been conducted to explore the various techniques used for knowledge reuse within organizations (Sandkuhl & Smirnov, 2018).
Hypothesis 4. Private Knowledge (PVRKM) positively influences the intention to predict innovation for supply chain financing.Hypothesis 5. Public Knowledge (PUBKM) positively influences the intention to predict innovation for supply chain financing.

Facilitating Conditions
Yabutani and Yamada (2023) identify the conditions required to motivate residents to engage in community activities, considering their individual characteristics.In addition, Nuseir & Elrefae (2022) incorporated within the context are facilitating conditions, customer experience, and brand loyalty, all of which influence the utilization of social media marketing, ultimately enhancing consumerbased brand equity.Also, facilitating conditions significantly impact students' intention to interact with one another (Wut et al., 2022).While the influence of facilitating conditions on the success of information system implementation is crucial, there is a lack of empirical research in the literature concerning the relationship between facilitating conditions and continuance intention in private higher learning institutions (Kamarozaman & Razak, 2021).Furthermore, the predictive role of facilitating conditions, perceived ease of use, and perceived usefulness in Iranian EFL learners' perceptions of mobile-assisted language learning (MALL) were examined by Ebadi and Raygan (2023).In addition, facilitating conditions emerged as the primary factor influencing their dynamic mathematics software usage behavior (Yuan et al., 2023).
Hypothesis 6. Facilitating Conditions (FCON) positively influence the intention to predict innovation for supply chain financing.Hypothesis 7. Facilitating Conditions (FCON) positively explain the innovation use behavior.

Innovation
Numerous studies have demonstrated that technological innovation has significantly reduced energy intensity (Wen et al., 2023).According to Li et al. (2023), the integration of digital finance with conventional finance and information technology (IT) holds great importance as it opens up new opportunities for green technology innovation and transformation within polluting industries Also, Chen and Liu (2023) used an innovation Spatial analysis, spatial Durbin models, and other methods are employed to analyze collaborative innovation between the logistics industry and manufacturing industry.Also, Irimia-Diéguez et al. ( 2023) focus on predicting FinTech innovation adoption, specifically examining the mediator role of social norms and attitudes in the innovation process.Finally, the relationship between embedding digital technology innovation networks and innovation behavior remains to be clarified and requires further investigation (Ge et al., 2023).Figure 1 shows the theoretical model for the study based on the literature.

Sample and Data
Primary data was collected from a Brazilian FinTech in the digital services sector in Rio de Janeiro, Brazil.The total valid data from the online questionnaire application was 124 respondents.The respondents work directly in digital services (digital accounts).The variables are defined from valid models of technology adoption and innovation prediction.
The sampling period for the collected data is from May 10 to June 30, 2023.The number of training and test samples is controlled to investigate prediction in the financial sector innovation context.The number of training cases was 96 (77,4%), and 28 (22,6,0%) cases were used for the test.The sample is a contextual benchmark, considering a specific time of data collection.A 7-point Likert scale was used (i.e., strongly disagree, slightly disagree, disagree, neutral, slightly agree, agree, strongly agree).

Procedures of Data Analysis
The structural equation modeling (SEM) (Hassan & Soliman, 2021) technique was used to analyze the data, with the partial least squares (PLS) (Hair Jr et al., 2020) method and ANN.The software used for the structural equation modeling analyses was "SmartPLS 4" (Ringle, Wende, & Becker, 2015), and for the neural networks (ANN), we used the Statistical Package for the Social Sciences (SPSS).The ANN used was the Multistrate Perceptron (MLP) type.
The following steps were applied for the structural equation modeling analysis: • Internal consistency and convergent validity were assessed using Cronbach's alpha and Composite Reliability (CR), respectively (Fornell & Larcker, 1981;Hair Jr. et al., 2009).• Three tests were used to analyze the discriminant validity of the model.The first was the Cross Factor Loadings (CFC), which is the correlation of the Observed Variables (OV) with the Latent Variables (LV) (Ringle; Silva; Bido, 2014).Another test was the Fornell-Larcker Criterion, which compares the square roots of the SEMs with the Pearson correlations (Fornell;Larcker, 1981).The third test was the Heterotrait-Monotrait Ratio Criterion (HTMT), confirmed by the Bootstrapping method, which is a more efficient criterion than Fornell-Larcker and is a kind of estimation of the correlation between the Latent Variables (Netemeyer;Bearder;Sharma, 2003).• Five tests were used to assess the criteria for evaluating the structural model.The Variance Inflation Factor (VIF) Collinearity Assessment verifies the existence of strong correlations between the variables, indicating collinearity problems (Hair et al., 2017).The Effect Size (f2) assesses the usefulness of each endogenous variable for adjusting the model (Cohen, 1988;Hair et al., 2014).
For the application of artificial neural networks (ANN): • Neural network analysis (ANN) was used to complement the findings of PLS-SEM in capturing non-linear links (Lee, Hew, Leong, Tan & Ooi, 2020;Wong, Tan, Ooi, Lin & Dwivedi, 2022) due to the limitations of PLS-SEM, which can only identify corrective and linear investigations (Lim, Lee, Foo, Ooi & Wei -Han Tan, 2021).

MEASURES Variables Used in the ANN Model
The variables used in the model are described in  (Venkatesh, 2006, 2022) (Figueiredo et al., 2023;Figueiredo & Ferreira, 2020).All variables are of scalar type.

Model Specification
This section discusses how conceptual modeling Table 2 was developed from two scientific models, the UTAUT Model and the SPINNER Model, applied respectively to technology adoption and innovation prediction.It presents the variables associated with the model estimated using SEM and ANN.The traditional UTAUT Model is known for evaluating the adoption of new technologies, and the SPINNER Model is known for its application in predicting innovation in SMEs.The results of the association between the two models are presented throughout the study, showing the gain of effectiveness in the integration.The first model was SEM, which was used to explain the antecedents of Behavioral Intention to Predict (BINP) (Hair Jr. et al., 2009).In addition, to assess convergent validity, the analysis of the model's internal consistency through Cronbach's alpha was performed, and the composite reliability (CR), including the average variance extracted (AVE) (Soliman et al. 2021).Considering the discriminant validity of the model, the average variance derived from the individual indicators was compared with the shared variance between the variables, including the Fornel-Larcker and HTMT criteria for the proposed model (Fornell & Larcker, 1981).The model evaluation was performed using variance inflation factor (VIF), coefficient, effect size indicator values (f2) or Cohen's indicator, model explanation coefficient R2, and predictive validity (Q2) or Stone-Geisser indicator.Table 2 shows the structural coefficients used in the model SEM for the dependent variable Innovation (Inn), with the Performance Expectancy (PEXP), Effort Expectancy (EEXP), Social Influence (SIN), Private Knowledge (PRVKM), Public Knowledge (PUBKM), Facilitating Conditions (FCON) and Behavioral Intention to Predict (BINP).This model was built based on Structural Equation Modeling, whose parameters were estimated using the SmartPLS® program.
The second model, ANN, was used to train and test the model Tabachnick & Fidell, (1991).The model explains the network training sample was equivalent to 96 (77.4%), while the test sample was 28 survey participants (22.6%).A total of 124 elements were considered valid.However, the processing in each neuron was done by the hyperbolic tangent and identity activation functions, as they are standard functions.In the tests performed with the other functions, these were the ones that presented the lowest root mean square error (RMSE) for training and testing.

Measurement Model (Internal Consistency and Convergent Validity)
The result of Cronbach's alpha indicates that all variables are within the acceptable range, demonstrating the reliability of the questionnaire (Table 3).It is recommended that Cronbach's alpha is greater than 0.7 (Hair Jr. et al., 2009), ranging from 0.791 to 0.944.The AVE value is acceptable because it has reached an average value greater than 0.5 and Composite Reliability (CR) (>0.7), according to Fornell and Larcker (1981) and Hair Jr. et al. (2009).

DISCRIMINANT VALIDITy (CROSS LOADINGS)
Table 4 presents the cross-correlation coefficients assessing the variables' discriminant validity.All the variables analyzed return a higher correlation with their latent variables of origin in relation to the other latent variables of the model.Thus, it can be concluded that the results meet the parameters defined for discriminant validity analysis (Henseler et al., 2015).

FORNELL-LARCKER AND HETEROTRAIT-MONOTRAIT RATIO CRITERIA
Two approaches were used to confirm discriminant validity, i.e., the Fornell-Larker (F-L) discriminant validity assessment and the Heterotrait-Monotrait ratio (HTMT) criterion using the bootstrapping method) (see Table 4).For the F-L criterion, the square root of the AVEs is larger than the correlations of the other variables (r ij for i ≠ j).As for the HTMT criterion, using the bootstrapping method for 5,000 subsamples, it is observed that the upper limits for 95% confidence are less than 1, except for the PEXP variable, which presented a value of 1.010 concerning the FCON variable.As the two criteria had their assumptions confirmed, the model presents convergent validity (Table 5).

Model Assessment (Variance Inflation Factor, Coefficient, Effect, and Predictive Relevance)
Table 6 shows the multicollinearity through the Variance Inflation Factor (VIF), whose values must be less than 5, thus confirming the non-multicollinearity between the variables (Hair et al., 2017).
The values of the effect size indicator (f 2 ) or Cohen's d indicator, the model explanation coefficient R 2, and the predictive validity (Q 2 ) or Stone-Geisser indicator are presented in Table 7.
Regarding the effect size values (f 2 ) presented in Table 8, it is considered that the BINP → INN effect presents a significant value (p < 0.05); that is, some of the other relationships may not be confirmed the relationship between the variables (Hair Jr et al., 2017), while the FCON → INN effect was not significant (p>0.05).The degree of explanation of the independent variable Innovation use Behavior (INN) in the model R 2 = 0.733 with p<0.05, while for the variable Behavioural Intention to Predict (BINP) was R 2 = 0.572 with p<0.05), considered as strong effect (Cohen, 1988;Hair Jr et al., 2017).
Another indicator of model fit quality is the Predictive Relevance or Stone-Geisser indicator, where Q 2 > 0 is indicative of predictive relevance; 0.01 ≤ Q 2 ≤ 0.075 weak degree, 0.075 < Q 2 ≤ 0.25 moderate degree and Q 2 > 0.25 strong degree (Chin, 2010;Hair Jr. et al., 2017).The Q 2 for the independent variables Behavioral Intention to Predict (BINP) and Innovation use Behavior (INN) presented Q 2 = 0.496 and Q 2 = 0.577, respectively, demonstrating predictive relevance with a strong degree (Chin, 2010;Hair Jr et al., 2017).Table 8 shows the results obtained between the latent variables in the model.
Figure 2 shows the PLS-SEM model.

MULTILAyER PERCEPTRON NETwORK
Table 9 shows the RNA case processing summary, and Table 10 shows the network information dealing with its construction characteristics.Excluding the Bias Unit The seven independent variables form the input layer or covariates: Performance Expectancy (PEXP), Effort Expectancy (EEXP), Social Influence (SIN), Private Knowledge (PRVKM), Public Knowledge (PUBKM), Facilitating Conditions (FCON), and Behavioral Intention to Predict (BINP).The covariates were rescaled by the standardized method, in which the mean is subtracted and divided by the standard deviation.The hidden layer contains unobservable network nodes.In this study, we worked with a hidden layer whose activation function was the hyperbolic tangent characteristic of using arguments with real values and transforming them in the interval (-1,1).The rescaling method was standardized, and the activation function was identity, i.e., it uses real values and returns them identically.To measure the quality of the predicted ANN, the sum of squares was used as the error function.
The predicted values in the input layer were treated in a standardized way, so their values oscillate between -1 and 1, while the values of the output layer were treated by the identity function, thus representing the determined association values (Table 11).To ensure that the constructed model can be used in other opportunities, there is the possibility to save and store the 'trained' ANN.
Table 12 shows that the error of the predicted model is 0.104, that is, 10,4% relatively low, which evidences that ANN can be used to predict the dependent variable innovation safely.The validation of the ANN estimated in this study is presented in Figures 3 and 4, which show the behavior of the predicted value for each observed value and the graph of the value of the residuals for each predicted value of the dependent variable, respectively.The behavior of the predicted values by the observed values is expected to present linearity.
In Figure 1, the analysis of the residuals shows that the normality hypothesis was met since the graph shows a behavior around the horizontal line centered at zero without characterizing a positive or negative trend.In Figure 2, the analysis of the residuals shows that the normality hypothesis was met since the graph shows a behavior around the horizontal line centered on zero without characterizing a positive or negative trend.
Table 13 and Figure 5 show information regarding the analysis of the importance of the independent variables in the construction of the ANN.It is observed that the variable Behavioral Intention Predict (BINP) is the one that contributes most to the prediction of RNA, that is, 100%,

DISCUSSION
The results support hypotheses 4, 5, 7, and 8 that the mixed model meets the prediction of the innovation behavior in the digital services sector, FinTech's.For these reasons, the ANN predicts the intention to  innovate in the Brazilian market.First, 74.4% of the network is oriented to training the model (with a relative error of 0,292), while 29.0% is oriented to testing the model.The ANN case processing summary and the network information deal with its construction characteristics.Second, we observed that the mixed methodological approach, SEM, and ANN could help to find the linear and non-linear relationships better, being that the error of the predicted model is 0.104, that is, 10.4% relatively low, which evidences that ANN, can be used to predict the dependent variable  to the study was 100%, followed by Performance Expectancy (PEXP) and Facilitating Conditions (FCON), both with more than 80%.
Considering hypothesis 4, our study supports that PVRKM (Private Knowledge) positively influences the intention to predict innovation for supply chain financing.This corroborates with Emtehani et al. (2023) that a comprehensive SCF framework by introducing two novel coordinating contracts that leverage trade credit financing was designed to address different problem settings within the supply chain.This is an example of an internal solution provided according to internal knowledge (Private Knowledge) to support the innovation process.
In terms of hypothesis 5, PUBKM (Public Knowledge) positively influences the intention to predict innovation for supply chain financing; our findings are consistent with Taleizadeh et al. (2022) in that the emergence of online peer-to-peer (P2P) lending platforms has introduced a promising FinTech business model that connects investors with capital recipients within supply chains (SCs).It means that the model is open to receiving external (public) knowledge to predict the sector's innovation through a digital platform.
Hypothesis 7, FCON (Facilitating Conditions), positively explains the innovation use behavior; according to Tsai (2023), advanced technology (blockchain) has been used to address current challenges in supply chain financing processes.This means that academics and professionals urgently need accurate and up-to-date knowledge of FinTech (Phan et al., 2022).In addition, Long et al. (2023) reinforce that to investigate how financial literacy and behavioral traits affect the adoption of electronic payment (ePayment) services, it was necessary to have an instrumental variable approach to facilitate the conditions to innovate.
Bringing hypothesis 8, INN (Innovation) positively influences innovation use behavior; it supports our empirical results according to Kao et al. (2022), where numerous financial institutions are keen on embracing technological solutions to bolster operational efficiency in managing SCF.This intricate process involves multiple participants and encompasses many complex financial activities.To conclude, our study brought value to the FinTech company, showing a mixed model that could be applied to predict the intention to innovate in the financial sector.Furthermore, the model showed accuracy using 50% of the proposed variables, being part of the UTAUT model and part of the SPINNER model, creating a balance in terms of hypotheses proposed and tested.

THEORETICAL IMPLICATIONS
The study results from applying the mixed method provide important theoretical and practical implications.Theoretically, we can extend the theory on digital services from FinTechs by considering some determinants.First, we understand which variables determine the intention to adopt innovation in FinTech's digital services by applying the hybrid model.Previous literature points to the use and relevance of the UTAUT model in several sectors, with greater relevance in the technology sector regarding technology adoption.In the case of the SPINNER model, its application in predicting innovation in SMEs was perceived.Second, we found that the behavioral determinant affects the intention to predict innovation (BINP) by 100%, regardless of the digital financial service developed by FinTech.Third, our findings advance theories on behavioral biases in decisions related to financial innovation in general and in technology adoption and digital service innovation prediction of FinTechs.In general, theories related to FinTechs broadly consider the context of investment and risk and avoid considering behavioral factors.

PRACTICAL IMPLICATIONS
This study has significant practical implications for startup entrepreneurs, FinTechs, and other financial sector institutions looking to deliver digital innovation products.This initial approach reinforces the importance of knowing and understanding the determinant variables of behavior related to financial innovation.
Innovation in the financial sector requires the consideration and knowledge of certain innovationrelated behaviors, such as the relationship between effort expectancy and social influence, which may constitute a barrier to adopting innovation in FinTechs.On the other hand, knowledge of the relationship between public and private knowledge can mediate innovation in the financial sector.In addition, knowledge of the enabling conditions, the expectation of effort, and especially the intention to foresee innovation can be perceived as facilitators of the innovation process of FinTechs.In conclusion, including behavioral programs can help improve innovation prediction in FinTechs.Contributing a set of internal and external knowledge to the business can mitigate behavioral resistance to innovation.This, carrying out a co-creation process and interacting with different financial system actors to develop intensive digital solutions.
This study aims to verify the application of determinants for SCF in Brazil.The accuracy of the mixed model, SEM, and ANN can help to find the linear and non-linear relationships better, being that the error of the predicted model is 0.104, that is, 10.4% relatively low, which evidences that ANN can be used to predict the dependent variable innovation safely.The application of the mixed model allowed significant results to be identified.The number of training and test samples is controlled to investigate prediction in the financial sector innovation context.The number of training cases was 96 (77,4%), and 46 (29,0%) cases were used for the test.
Although this study found that the mixed approach, including SEM and ANN, can help to find the linear and non-linear relationships better, reliably predicting the dependent variable innovationthis research shows some limitations.First, even though ANN is considered an excellent data analysis method, it is a model that finance professionals may probably find difficult to apply.Second, data from a FinTech based in Brazil was used in the study.Although the FinTech market in Latin America has been growing exponentially, influenced by the advancement of digitalization, the results cannot be generalized.The result of the study cannot represent a global view of the sector, as it lacks a more detailed and extended analysis to address a global market in terms of financial innovation.
Future studies can consider a few approaches: First, they can test the variables with new statistical approaches and compare them with the current model.Second, they can consider other digital services, regions, and countries.Third, researchers can compare the accuracy of other methods or mixed models in predicting innovation.Fourth, future studies can consider samples from various markets and compare the results against a single market or digital service.Finally, another future suggestion is exploring which artificial neural network approaches can be used to predict predictive behavior better.Figueiredo, R., Soliman, M., Al-Alawi, A. N., & Fatnassi, T. (2023).Could the 'Spinner Innovation'and 'Triple Helix'Models Improve System Innovation? Applied System Innovation, 6(2), 42. doi:10.3390/asi6020042 Hypothesis 8. Innovation (INN) positively influences innovation use behavior.

Figure
Figure 1.Theoretical model

Figure 3 .
Figure 3. Graph of predicted value by observed value

Figure 4 .
Figure 4. Graph of residuals per prediction

Table 2 . Analysis of structural coefficients
Note: SD = Standard Deviation

Table 8 . The structural model and hypotheses testing
* Standard Deviation

Table 12 . Model summary
Dependent variable: Inn; a: Error calculations are based on the test sample.

Table 13 . Importance of the independent variable (innovation) Variable Importance Normalized Importance (%)
innovation safely.The importance of innovation variables' Behavioral Intention to Predict (BINP)