Spatio-temporal Prediction under Scarcity of Influencing Variables: A Hybrid Probabilistic Graph-based Approach

• Published in: 2017 Ninth International Conference on Advances in Pattern Recognition (ICAPR) pp. 1 - 6
• Main Authors: Das, Monidipa, Ghosh, Soumya K.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.12.2017
• Subjects: Bayes methods; Computational modeling; Fuzzy Bayesian network; Hybrid probabilistic model; Predictive models; Probabilistic logic; Residual correction; Spatio-temporal prediction; Time series analysis; Training; Uncertainty
• DOI: 10.1109/ICAPR.2017.8593054

Spatio-temporal (ST) prediction is one of the major families of spatio-temporal data mining, which aims at learning a model that can predict the target variable by using a given set of explanatory variables in a spatio-temporal framework. It has enormous application in various domains, including environmental management, transportation, epidemiology, climatology and so on. However, a major issue in ST prediction of any variable is the unknown factors that can have significant influence on it, or the unavailability of the data on the factors that influence the prediction variable. In such cases, due to lack of appropriate data on influencing factors, the effectiveness of many of the existing space-time prediction models, especially those based on causal graph-based approach, are significantly hindered. In the present paper, an attempt has been made to address this issue by proposing a hybrid probabilistic model based on fuzzy Bayesian network with incorporated residual correction mechanism (FBNRC). The incorporated fuzziness aids in reducing uncertainty during parameter learning process, whereas the added functionality of residual correction helps to compensate for the unknown factors while generating inference on the prediction variable. Experimentation has been carried out on spatio-temporal prediction of meteorological time series in the state of Delhi, India. The results of prediction are found to be encouraging.