Event Predictability: A Uniform Form for IoT-Based Nondeterministic Social Systems

Abstract

Integrating massive social data with traditional social sciences, the Computational Social Science (CSS) is crucial for understanding the Internet of Things (IoT)-based nondeterministic social systems. Event predictability is the fundamental premise of widespread societal event predictions with CSS. Due to data quality, model suitability and the nondeterministic nature of IoT-based social systems, the event predictability is difficult to be characterized. Based on Turing computability and prediction error tolerability, this paper posits a uniform event predictability theory. With discrepancy and Rademacher complexity, the generalization error bound is utilized to represent data quality and model suitability. Together with thresholds for the generalization error bound and confidence, the event predictability is modeled in a probabilistic manner to capture the nondeterminism within IoT-based social systems. The event predictability theory is theoretically proved and validated, and utilizing the proposed Approximation Algorithm for Discriminating Event Predictability (AADEP), its applicability is further verified by experiments on a real-world dataset for IoT-based nondeterministic social systems.