Development of Model to Predict Natural Disaster-Induced Financial Losses for Construction Projects Using Deep Learning Techniques

Abstract

This study goals to develop a model for predicting financial loss at construction sites using a deep learning algorithm to reduce and prevent the risk of financial loss at construction sites. Lately, as the construction of high-rise buildings and complex buildings increases and the scale of construction sites surges, the severity and frequency of accidents occurring at construction sites are swelling, and financial losses are also snowballing. Singularly, as natural disasters rise and construction projects in urban areas increase, the risk of financial loss for construction sites is mounting. Thus, a financial loss prediction model is desired to mitigate and manage the risk of such financial loss for maintainable and effective construction project management. This study reflects the financial loss incurred at the actual construction sites by collecting claim payout data from a major South Korean insurance company. A deep learning algorithm was presented in order to develop an objective and scientific prediction model. The results and framework of this study provide critical guidance on financial loss management necessary for sustainable and successful construction project management and can be used as a reference for various other construction project management studies.

Predicting financial losses due to apartment construction accidents utilizing deep learning techniques

“…The MRA model was established using the IBM Statistical Package for the Social Sciences (SPSS) version 23.Multiple regression analysis is a regression analysis method that estimates the fundamental correlation between variables utilizing a statistical method. It is generally agreed in the field of prediction59 . As a result, paralleling result values of the DNN model and the MRA model, the DNN model exhibited a lesser prediction error rate of 41.3% in MRA and 46.0% in RMSE than the MRA model.…”

“…With the purpose of developing an optimal model for learning input and output variables, network scenarios and hyper-parameters were firm over and done with trial-and-error methods. 59,60 .…”

A deep neural network algorithm-based approach for predicting recovery period of accidents according to construction scale

“…To achieve an optimal model configuration, it is crucial to establish the appropriate number of nodes and layers in the network structure and set hyperparameters such as dropout rate, batch size, epoch count, choice of optimizer, and activation functions. Hyperparameters are parameters that control the learning process [ 48 ]. For example, dropout is a regularization penalty to solve the overfitting problem that causes poor performance of deep learning models.…”

Earthquake intensity estimation via an artificial neural network: Examination of different network designs and training algorithms

“…Bu bölümde literatürde yer alan çeşitli tahmin çalışmaları Tablo 1'de görüldüğü gibi özetlenmiştir. 32], Rassal ormanlar [19,22,32], Regresyon modelleri [27], Yapay sinir ağları [13-15, 16-20, 30-32], Diğer [19,22,23,28] Afet sonrası oluşan hasar Derin öğrenme [50,51], Destek vektör makineleri [52], Karar ağaçları [48,52], Rassal ormanlar [41,47,48], Regresyon modelleri [41,49,52], Yapay sinir ağları [34, 36-38, 41, 45, 47], Diğer [35,39,40,42,43,46,48] olasılığı ve sınırlı büyüklük aralığında bir depremin meydana gelme olasılığı olmak üzere iki tip tahmin yapılmaktadır. Corbi vd.…”

“…[49] tarafından sunulan regresyon teknikleri ve Kim vd. [50] tarafından derin öğrenme kullanılarak yapılan çalışma örnek olarak gösterilebilir. Bazı çalışmalarda ise afet kaynaklı yeryüzünde oluşacak hasara dikkat çekilmiştir.…”