A Distributed Data-Driven and Machine Learning Method for High-Level Causal Analysis in Sustainable IoT Systems

Abstract

A causal relationship forms when one event triggers another's change or occurrence. Causality helps to understand connections among events, explain phenomena, and facilitate better decision-making. In IoT systems, massive consumption of energy may lead to specific types of air pollution. There are causal relationships among air pollutants. Analyzing their interactions allows for targeted adjustments in energy use, like shifting to cleaner energy and cutting high-emission sources. This reduces air pollution and boosts energy sustainability, aiding sustainable development. This paper introduces a distributed data-driven machine learning method for high-level causal analysis (DMHC), which extracts general and high-level Complex Event Processing (CEP) rules from unlabeled data. CEP rules can capture the interactions among events and represent the causal relationships among them. DMHC deploys a two-layer LSTM attention mechanism model and decision tree algorithm to filter and label data, extracting general CEP rules. Afterward, it proceeds to generate event logs based on general rules with heuristic mining (HM), extracting high-level CEP rules that pertain to causal relationships. These high-level rules complement the extracted general rules and reflect the causal relationships among the general rules. The proposed high-level methodology is validated using a real air quality dataset.