A cost model to improve short-term underinsurance of residential buildings in South Africa

This thesis demonstrates the development of a case-based reasoning (CBR) enabled cost estimating method for residential buildings in South Africa for application in the insurance environment to address the continuous under-insurance gap perpetuated by inappropriate cost models. The CBR comprises the four steps of retrieving, re-using, revising and retaining cases from the custom-designed dataset. The dataset contains data for forty-five cases based on traditional building elemental estimates and fourteen design features. The elemental estimates are based on the built environment’s entrenched measuring methodology, and the features are designed to address shortcomings in the currently applied cost models for determining replacement cost estimates for insurance purposes. Estimates based on these measuring methods are still regarded as the most accurate predictions of actual cost. The measuring process is laborious, time-consuming, requires specialist-built environment involvement, and is costly. The cost outweighs the perceived risk of insuring for the correct sum. The proposed CBR method addresses all these aspects. The k-nearest neighbour (kNN) machine learning algorithm performs the first step to retrieve the cases from the dataset with features most similar to the case under investigation. The other steps of re-using, revising and retaining are performed through mathematical model-based reasoning. The mathematical estimating model requires the input of fourteen design features extracted from the case under investigation’s drawing that are pro-rated to the features of the retrieved nearest neighbours and multiplied by the elemental values to produce replacement cost estimates for the case under investigation. One hundred and thirty-five estimation iterations based on the chosen nearest neighbours were performed. The model shows the promise to provide accurate replacement cost estimates for insurance purposes, as the results obtained show 59% of the iterations to be within 10% accuracy of the elemental estimates. Machine learning techniques are not widely practised in cost modelling in South Africa’s built environment. The potential for developing and implementing cost models for various purposes, more than just insurance purposes, is immeasurable and could place the built environment truly on the Fourth Industrial Revolution trajectory.