Trend Events and Lag Coefficients in Time Series Association Rule Discovery

Abstract

Trend events reveal long-term changes in time series data and help capture long-term pattern shifts. Lag coefficients, on the other hand, focus on the impact of past events on the present or future, helping to uncover time delays and causal relationships. In time series association rule discovery, considering trend events and lag coefficients can enhance the model's predictive power and provide more accurate decision support, particularly in fields such as finance, market analysis, and epidemiology. Both of these factors are key elements for effective modelling and rule mining, and are crucial for extracting meaningful time series patterns. This paper analyses those key challenges and introduces a new approach to solve these problems by constructing trend events and solving for lag coefficients. Based on this approach, some algorithms are proposed to improve classic association rule discovery method. The experiment with PM2.5 data set proves the effects of discretization of trend event and shows the hidden frequency pattern considered lag coefficient. This research represents a meaningful attempt to improve the effect of association rules discovery within multiply time series.