Prediction models for structural deterioration of urban drainage pipes using ordered probit and neural-fuzzy techniques

Predicting structural deterioration of drainage pipes for stormwater runoff is an important part of drainage maintenance programs. The structural deterioration is affected by various factors such as pipe size, age and soil type, among other variables and is defined using three condition states, with one being good, two being fair and three being poor over the lifetime of pipes. In this paper, two prediction models using ordered probit (OPPM) and neuro-fuzzy (NFPM) are developed for predicting the structural condition of urban drainage pipes. These two models are compared against each other by using the Goodness-of-fit test and two scalar performance measures, namely, overall success rate (OSR) and false negative rate (FNR). The predictive performance of OPPM might be affected by the noisy data owing to the nature of its statistical structure. The noisy data are inherent with the condition monitoring and assessment of the structural deterioration (i.e. collapsing into three condition states) and the vagueness of the input factors. This is what the adaptive neuro-fuzzy inference system (ANFIS) reportedly can handle well. The ANFIS is based on two powerful artificial intelligence techniques, multi-valued logical system (fuzzy logic) to account for noisy data and neural networks to map input factors to accurate outputs (i.e. structural condition). A case study was used to demonstrate the applicability of OPPM and NFPM. The results showed that the NFPM was more suitable for modelling structural deterioration of stormwater pipes as substantiated by the Goodness-of-fit test. The NFPM consistently outperformed the OPPM in the train and test data set, however while these results are promising, further improvement is required before it can be used as a predictive model without additional filed expert opinion and confirmation.