An Overview of Pavement Degradation Prediction Models

Table 2

Pavement performance prediction studies in shallow machine learning.

Technique used	Pavement indicator	Data sources	Metrics	References	Strength	Weakness
Artificial neural network ANN	IRI	LTPP	R^2, RMSE, MAE, MSE, CF, and VAF	Abdelaziz, et al. [64]	Able to work with vast amounts of data and most challenging problems, change the structure to the used parameters, suitable for time-series problems.	Expensive to train, requires long training time and massive data
		Observed	R^2, RMSE, MAE	Lin et al. [50], Mallika, et al. [65]
	PCI	Observed	MAE, RMSE, and R^2	Shahriazari et al. [46], Jalal, et al. [47]
Neuro-fuzzy Model NFM	IRI	LTPP observed	R^2 and RMSE correlation factor R	Soncim et al. [66], Ngnyen, et al. [32]	Suitable for complex data interactions, easy to scale and have high converge	Requires huge data, complex and difficult to debug.
Regression	IRI	LTPP	R^2, MSE, RMSE	Elhadidy et al. [22], Piryonesi and El-Diraby [29]	Simple, requires a minimum number of parameters, suitable in classification and recognition works	Expensive, not able to work with a multifeatures dataset and poor in presenting the extreme events.
	PCI	Observed	R^2	Ahmed, et al. [63]
Support Victor machine	IRI	Observed	Ransom output error	Roberts and Attoh‐Okine [57]	Training is simple and relatively easy, suitable in high-dimensional data	Requires high memory and more time for training the model.
	IRI	LTPP	MSE MAE and RMSE	Kargah-Ostadi and Stoffels [67]

R²: coefficient of determination, RMSE: root mean squared error, MAPE: mean absolute presenting error, CF: correction factor, VAF: variance account for, MAE: mean absolute error, RE: relative error, MSE: mean squared error, and SDMSE: standard deviation of mean squared error.