| description abstract | Age has often been the only factor considered in predictive models of degradation. This study, however, assesses the influence on degradation of coastal exposure for concrete beams, level of utilization for rendered cement floors, and rainfall for timber windows. First, the difference between random data and data categorized on the basis of high or low levels of environmental factors was explored to establish whether they had a perceptible influence on degradation. Next, five types of models were explored for fitting the data and making predictions: namely Markov chain, multiple linear regression, simple neural network, deep neural network, and random forest. Among the environmental factors, coastal exposure on concrete beams had the greatest influence, while rainfall on timber windows the least. Random forest modeling was the most accurate and was also explored using the local interpretable model-agnostic explanation (LIME) technique, which revealed that the environmental factor effects were more evident during the mid-life of elements rather than at the early or late stages. Including environmental factors in degradation models in addition to element age will increase their accuracy and portability. | |
