Table 2 AI for forecasting Type of AI... | Emerald Publishing

Table 2

AI for forecasting

Type of AI	Use/type of forecasting	Industry	Reference	Limitation
Neural networks + supervector machines	Predict project performances (cost and schedule)	Construction industry	Wang et al. (2012)	ANNs and SVMs act as “black boxes,” their decision-making processes are not transparent; SVMs are effective in classification tasks but they can struggle with generalizability when applied to new projects that differ from those in the training set
Neural networks + Support vector machines	Improve Cost and Duration Prediction Accuracy	Construction industry	Darko et al. (2023)	The introduction of Deep Neural Networks (DNN) and Support Vector Regression (SVR) introduces complexity in terms of model configuration, training, and optimization
Neural networks	Risk prediction in tunnel construction frastructure	Luo et al. (2024)	Feature selection does not consider interrelationships between variables
Neural networks	Predict construction cost of large sport field facilities	Construction industry	Juszczyk et al. (2019)	Do not possible to update on different time frame the database, limiting analysis effectiveness
Neural networks	Predict waste generation rate of building demolitions	Construction industry	Cha et al. (2023)	ANNs are sensitive to the input data variations and might not perform well if the data is not representative of the typical scenarios encountered during demolition projects; ANN requires accurately labeled data for training
Neural networks	Predict project success	Construction industry	Ko and Cheng (2007)	The model used Fuzzy Lofic, Neural Networks and Genetic Algorithm. It requires extensive computational resources and time for training and optimizing hyper-parameters through Bayesian inference and Particle Swarm Optimization
Neural networks	Cost and time forecasting of megaprojects	Megaprojects and Infrastructure	Natarajan (2022)	The method cannot quantify all the projects risks and uncertainties; Outliers megaprojects cannot be predicted
Long short-term memory neural networks (LSTM) + ARIMA and ARIFMA	Predict the Volatility of Highway Construction Cost Index	Megaprojects and Infrastructure	Cao and Ashuri (2020)	If the change is in the testing period, ARIMA and ARIFMA can only detect the periodic ones, and cannot catch unhappened ones. If the change is in the training period, the time series model is insensitive to it when change hap-pens near the end of the training sample or distant to the end

Type of AI	Use/type of forecasting	Industry	Reference	Limitation
Neural networks + supervector machines	Predict project performances (cost and schedule)	Construction industry	Wang et al. (2012)	ANNs and SVMs act as “black boxes,” their decision-making processes are not transparent; SVMs are effective in classification tasks but they can struggle with generalizability when applied to new projects that differ from those in the training set
Neural networks + Support vector machines	Improve Cost and DurationPrediction Accuracy	Construction industry	Darko et al. (2023)	The introduction of Deep Neural Networks (DNN) and Support Vector Regression (SVR) introduces complexity in terms of model configuration, training, and optimization
Neural networks	Risk prediction in tunnel construction frastructure	Luo et al. (2024)	Feature selection does not consider interrelationships between variables
Neural networks	Predict construction cost of large sport field facilities	Construction industry	Juszczyk et al. (2019)	Do not possible to update on different time frame the database, limiting analysis effectiveness
Neural networks	Predict waste generation rate of building demolitions	Construction industry	Cha et al. (2023)	ANNs are sensitive to the input data variations and might not perform well if the data is not representative of the typical scenarios encountered during demolition projects; ANN requires accurately labeled data for training
Neural networks	Predict project success	Construction industry	Ko and Cheng (2007)	The model used Fuzzy Lofic, Neural Networks and Genetic Algorithm. It requires extensive computational resources and time for training and optimizing hyper-parameters through Bayesian inference and Particle Swarm Optimization
Neural networks	Cost and time forecasting of megaprojects	Megaprojects and Infrastructure	Natarajan (2022)	The method cannot quantify all the projects risks and uncertainties; Outliers megaprojects cannot be predicted
Long short-term memory neural networks (LSTM) + ARIMA and ARIFMA	Predict the Volatility of Highway Construction Cost Index	Megaprojects and Infrastructure	Cao and Ashuri (2020)	If the change is in the testing period, ARIMA and ARIFMA can only detect the periodic ones, and cannot catch unhappened ones. If the change is in the training period, the time series model is insensitive to it when change hap-pens near the end of the training sample or distant to the end

Source(s): Authors’ own creation