Conversational agents powered by artificial intelligence are transforming customer support. Many include artificial empathy (phrasing that validates user emotions) under the assumption that it improves satisfaction and organizational reputation, yet its effect during service failures remains unclear. This study compared an empathic and a neutral conversational agent in a standardized support scenario. In total, 84 participants completed a pre-test, 41 interacted with the agents, and 9 completed a post-test. Emotional state was measured with the Positive and Negative Affect Schedule (PANAS), perceptions with a Likert-scale battery, and AI literacy with the Meta AI Literacy Scale (MAILS). All 41 chatlogs were analyzed for behavioral markers of frustration, escalation, and resolution. The empathic agent resolved more cases (62
A smart metropolis promotes social inclusion through adaptive technologies that go beyond operational efficiency . This research applies Artificial Intelligence (AI) to smart city environments, proposing an accessible navigation system to redesign traditional urban infrastructures as genuinely inclusive spaces. The Guadalajara Zoo serves as a living laboratory to validate these solutions, focusing on children with motor disabilities and functional hearing impairments. The methodology implements Dijkstra's algorithm to calculate optimal, personalized routes, considering granular variables such as distance, slope, pavement type, obstacles, and the user's specific mobility profile. Computational simulation of the model successfully generated consistent itineraries for three accessibility profiles (wheelchairs, crutches, and strollers) within a 60-minute window. The results showed an optimized visit capacity ranging from 4 to 12 points of interest per hour, depending on the depth of the visit and the mode of mobility. In a typical visit scenario ( 5 min/stop), the optimized model allowed a child in a pushchair to visit 6 to 7 habitats in one hour. The system personalizes the experience by calculating the distance and total time based on the user's available time and needs, demonstrating an improvement of up to 25
This article presents a study on classifying white and red wines using the Decision Tree Machine Learning technique. The main objective is to reliably predict the color of the wine—white or red—using physical and chemical variables as inputs to machine learning models. A 10-fold stratified cross-validation was applied to two approaches: a base model with a Decision Tree (Decision-TreeClassifier) and another with an additional stage of nonlinear dimensionality reduction using KernelPCA. The best result was obtained with a five-dimensional k-fold reduction, achieving an accuracy of 99.1
Optimization in complex spaces requires algorithms capable of adequately balancing exploration and exploitation. This work proposes a hybridization of Ephemeral Collective Intelligence (MA) and Cockroach Infestation (RIO) algorithms by incorporating a fuzzy system that allows switching between the two algorithms according to the needs of the search process. This system dynamically regulates the switching and execution intensity of the algorithms based on iteration and population diversity, allowing MA to focus on refining the most promising regions of the search space, while RIO introduces the necessary diversity to prevent premature convergence. Experimental results obtained on 15 benchmark mathematical functions demonstrate that the hybrid proposal with fuzzy control outperforms MA and RIO individually on 9 of the 15 functions evaluated. As future work, we propose incorporating additional metrics, such as robustness, to increase the accuracy and effectiveness of the fuzzy adapter.
Computer vision-based wildfire detection is a crucial task for enabling a rapid firefighter response, allowing to mitigate several damages in ecosystems and economy. Moreover, wildfire segmentation methods allow for determining the precise position where the fire is present within the image. A number of rule-based models have been proposed since they are human-readable. Moreover, the color content is the most utilized feature to characterize the fire pixels. Generally, rule-based models determine fire pixels in a binary manner, making it difficult to handle the uncertainty present in the data. In this research work, we propose a system for automatically calculating color-based rules using fuzzy operators to determine the fire pixel. Specifically, we propose to use calculated color-based features. In contrast to other rule-based models, which perform fire segmentation using inequalities, our approach constructs a fire segmentation model by handling uncertainty in a fuzzy approach. By using a genetic approach, the proposed system searches for the best ensemble of fuzzy rules to form the model. According to the results obtained, the proposed model outperforms the state-of-the-art models in various segmentation metrics.
