Human-Machine_Trust_Prediction_Using_Behavior_Measures_and_Trust_Relationships

We conducted an experiment with 13 participants using MATB tasks to collect trust-related data. Through data analysis, we show the validity of our experiment design and identify nine features with high correlations to trust for trust prediction. Using this dataset, our model outperforms a recent trust model and a classic trust model by at least 16%. The data contains original files with names of xxx-xxx-xxx.csv indicating MATB logs files and when they are generated, last_name.csv or last_name.txt indicating the eye tracking data of each individual of every round. The files with names of last_name.json are processed data. The keys in the json files are features, and the corresponding values are time series data of the features. We transform the time-series data into overlapping time slices, and the length of each depends on the desired sequence length. We determine the best sequence length with a comparing analysis discussed in our paper. The script file is an example of the transforming code.