Human–AI Interaction In Machine Learning-Based Intelligent Interfaces: A Design And Evaluation Study

Abstract

Human–Artificial Intelligence (AI) interaction in intelligent interfaces is crucial for developingcollaborative decision support between users and Machine Learning (ML) algorithms, particularly in applications involving smartwatches for healthcare. Nevertheless, the effects of interface design and interpretability techniques related to user trust, cognitive load, and decision quality in these types of applications have not been thoroughly investigated. The objective of the research is to design a human-centered intelligent interface and assess its interpretability, usability, and impact on decision-making performance, while increasing the accuracy of prediction results through a hybrid approach. The proposedGlow Worm Swarm Optimization- Dynamic K-NearestNeighbor (GWSO-DKNN) solution employs sensor data from smartwatches such as heart rate, movement, sleep, and activity information. Data preprocessing is performed using the Min-Max Normalization technique, whereas the Fast Fourier Transform (FFT) method is utilized for feature extraction in physiological time-series signals. An experiment is carried out to test multiple interface designs, while performance measures include accuracy, recall, F1-score, and log loss. The proposed GWSO-DKNN model demonstrates high precision of 96.77%, recall of 98.36%, F1-score of 97.56%, and low log loss of 5.62%, it can lead to improved user understanding. Both the development and evaluation of the proposed model are conducted within the Python programming environment. The proposed GWSO-DKNN classifier demonstrates higher performance compared to conventional classifiers like Random Forest (RF), Support Vector Machine (SVM), Naive Bayes, Logistic Regression (LR), and Perceptron.