**4. Discussion**

In general, for this study, consumer perception had higher scores for self-reported information for the control samples, which did not contain insects, and the lowest scores for full insects, as expected. These scores were similar to the sample with insects as an ingredient, in which no parts of insects were visible, which is in accordance with previous literature [1,14,31]. However, mid-range liking, FS Taste, and purchase intention were obtained for visible insects mixed with more familiar products, such as toast and avocado mash (Table S1).

When combining self-reported information with the subconscious and emotional responses through the PCAs (Figure 2), it was found that high and similar variability of data was explained for both the Asian and non-Asian participants (>70%). For the Asians (Figure 2a), more negative emotional responses were associated with insect-based samples compared to the controls, with a clear separation in the vertical plane (PC1). For the non-Asians (Figure 2b), a similar separation was observed for the samples with visible insects, with a difference in the insect tortilla chip. These results may seem contrasting with the self-reported data; however, the subconscious responses are related to the first impression of the samples for visible, aroma, and taste responses from consumers, which are spontaneous and more complex when tasting food and beverage products of any kind [32–34]. Similar relationships were found using the correlation matrix analysis (Figure 3) and the further MFA combining the self-reported data, biometrics, and CATA.

A deeper understanding of consumer acceptability, liking, and intention to purchase new insect-based food products is extremely important, since around 95% of new food products may fail in the market without proper assessment [35,36]. Predictive modeling incorporating cultural backgrounds may offer more information and the possibility of automation for the decision-making process or product variation when developing new insect-based food products. Recent research has been based on automatic estimations of liking based on facial expression dynamics, especially for infants, since self-reported data may not be easily obtained [37,38]. However, research on automatic assessments based on biometrics is rarer [15,28,32].

By using ML modeling considering the separation of cultural background, it was possible to obtain high accuracy (≥92%) in the prediction of liking based on the biometrics from Asians (Table 5; Figure 5a) and non-Asians (Table 5; Figure 5b) using the Bayesian Regularization ANN algorithm. However, since ML using ANN is considered to be a robust method to detect patterns within data, the cultural distinction could be an internal feature of a general model, as shown in Table 5 and Figure 5c. This general model, with an accuracy of 89%, may be used to detect the liking levels of snacks containing insects as part of their ingredients or as whole insects. Further research is required to test and model the biometric responses from a wider variety of insect-based food or beverages.

A quicker analysis for consumer acceptability, liking, and intent to purchase can be achieved as a first approach by presenting images of dishes prepared with insects through the BioSensory app to obtain biometrics from panelists. It has been previously shown for packaging assessments that presenting images of the packaging, and the physical packaging samples resulted in non–significant statistical differences in their appreciation by panelists in a sensory trial [39]. Similarly, for beer tasting panelists looking at videos of beer pouring gave similar levels of liking and general appreciation compared to those that tasted the same beers [15,40]. The applicability for the assessment of images of insect-based dishes is supported by the data presented in this study, especially related to the statistically significant correlations between the appearance and overall liking for both cultures (Figure 3a,b) and the higher variability of the data explained by PC1 for both Asians and non-Asians (Figure 2a,b).

The advantages of using both the self-reported and biometric responses rely on the fact that they allow us to obtain the first and subconscious reaction that consumers have towards the product they are assessing as well as the way they may modify their responses after the thinking process. This aids in a deeper assessment of consumers' behavior and acceptability towards food and beverages to understand the target market and develop products according to their needs. The results from Asians and non-Asians were statistically analyzed separately due to the cultural differences that may be related to the expression of emotional responses. This was also the main reason for developing ML models for each culture and a general model considering both cultures. However, despite that Model 3 (general model) had a slightly lower accuracy, it was shown that ANN is able to accurately find patterns among data that may be related to differences in responses from different cultural backgrounds to predict liking regardless.
