DescriptionWith the growth in mobile social networks, social internet of things, and cyber-physical-social systems, there is an ever growing need to model and understand human beings as they interact with other humans and socio-technical ecosystems. In this dissertation, we focus on modeling three core human concepts – trust propensity, altruism propensity, and interpersonal trust using mobile phone metadata. Traditional methods for understanding an individual’s propensities and behaviors have been surveys and lab experiments. However, the growth of “personal big data”, which includes the use of various personal ubiquitous devices, is allowing for human behaviors and propensities to be modeled via lower-cost, quick, automated methods. This dissertation proposes a new methodology to model human behaviors and propensities based on phoneotypes (phone-based observations of a combination of people’s traits) that aims to complement traditional methods like surveys with a ubiquitous data-driven automated method. The analysis and modeling employ multiple deep and shallow machine learning algorithms and are based on two datasets - Rutgers Well-being Study and MIT friends and family dataset. Overall, the findings suggest that: (1) many phone-based features are associated with participant’s altruism, trust, and interpersonal trust scores;
(2) phone-based prediction models for altruism, trust propensity, and interpersonal trust performed statistically significantly better than comparable demography-based models. This dissertation paves way to study the associations between human behavioral propensities and long-term “in the wild” socio-mobile behavior, and to utilize “personal big data” with shallow and deep machine learning approaches to model altruism, trust, and interpersonal trust. A better modeling approach for human beings will have multiple applications in fields like healthcare, well-being, and urban planning.