Skip to Main Content
The characteristic data applied for establishing predicting models of pipe failure in water distribution system (WDS) should be examined and screened in the first place. The effects of different way of data processing based on a series on maintenance records are compared, which includes a classification method by Bayesian theorem, a hypothesis testing by analysis of variance and an association rule of data mining. The result shows the Bayesian theorem is applicable for nonparametric analysis, and can give classification limits intuitively. But this method may neglect the relations between different variates and effects. It's proper to adopt this method when there are few pipe characteristics available. The analysis of variance (ANOVA) is able to point out the interaction of variates and deal with both parametric and nonparametric tests. The association rule can provide correlations between different attributes, but may also bring excessive and nonsense results and it appears that more detailed records are required to make use of this method. It is suggested that the data should be carefully examined in appropriate method before applied to a model.