Response-based methods to measure road surface irregularity: a state-of-the-art review

European Transport Research Review

An Open Access Journal

Table 6 Summary of ML methods for pothole detection and roughness index estimation

System name/by	Machine learning	Function		Approach			Additional
System name/by	Machine learning	PD	RE	C	F	S	GPS	Data	Crowd
Pothole Patrol [102]	Clustering + training detector	✔		✔
Nericell [103], TrafficSense [104, 105]	Z-peak method/ Clustering + training detector	✔				✔
PRISM [106]	Z-peak method + training detector	✔				✔			✔
[107]	supervised ML	✔				✔
P³ [108]	Clustering + training detector	✔				✔		✔
PADS [111]	K-mean clustering	✔		✔
BDS [112]	K-means clustering + RF	✔				✔
[113]	Naive Bayes algorithm + K-nearest-neighbor		Relative			✔
[114]	SVM + unsupervised ML	✔	Relative			✔
D&Sense [115]	SVM + DTW	✔				✔			✔
RoadMonitor [116], RoADS-based [117, 118]	SVM, SVM + SWT	✔				✔
[119]	SVM + FFT, cross validation	✔				✔	✔
[120]	SVM + WPD, feature selection	✔				✔
Wolverine [121]	SVM + K-means clustering	✔				✔
[122] [123]	SVM + data filter, sliding window, greedy forward feature selection	✔		✔			✔	✔	✔
VRNI [124]	SVM + filter, moving window, feature extraction	✔			✔
CRISP-DM-based [125, 126]	various algorithms comparison	✔	Relative			✔
[127, 128]	various algorithms comparison	✔	Relative	✔			✔	✔
RoadSense [129], Pothole Lab [130]	various algorithms comparison	✔				✔			✔
[131]	various algorithms comparison	✔				✔
[132]	various algorithms comparison	✔		✔				✔
CRSM [109, 110]	iGMM clustering	✔	IRI		✔			✔	✔
[133]	SVM + WPD, Random forest		IRI			✔	✔	✔	✔
[134]	ANN + feature selection		IRI	✔				✔
[7]	Fuzzy classifier	✔	Relative			✔		✔	✔

Relative: Pothole-based roughness index;
ANN, SVM, RF, DT: Artificial Neural Network, Support Vector Machine, Random Forest, Decision Tree;
PCA, WPD, DWT: Principal Component Analysis, Wavelet Package Decomposition, Discrete Wavelet Transform.