Predicting Ocean Salinity and Temperature Variations Using Data Mining and Fuzzy Inference

Global ocean salinity/temperature variations are attracting increasing attention, due to their influence on ocean-atmospheric changes and their potential for improved climate forecasting. The goal is to analyze historic salinity/temperature data to make predictions about future variations. Traditional statistical models that assume data independence are not applicable as ocean data are often inter-related. Association rules mining can be used to find interesting salinity and temperature patterns, however, the traditional method ignores spatial and temporal information in the data. This study proposes a strategy that employs inter-transaction association rules mining to discover salinity/temperature patterns where spatial/temporal relationships are considered. Next, a fuzzy inference is used to predict salinity/temperature variations. The fuzzy inference rules are derived from a set of inter-transaction association rules that are discovered from Argo data. The strategy is highly efficient as a reduced prefix-projected itemset algorithm with a small space and time complexity is employed in the search for large inter-transaction itemsets. This proposed strategy is unsupervised as it does not rely on domain experts for designing the fuzzy rule base. Experimental results demonstrate that the proposed strategy effectively predicts abnormal salinity/ temperature variations.