Table 7.

Machine learning (ML) and deep learning (DL) models and algorithms used by previous researchers in mobile phone data studies. The abbreviation of algorithms is shown in the back matter.

References	Algorithm/Model	Objective
[102]	SVM, NB	To classify user relationships
[8]	FCM	To classify urban land use in Singapore
[4]	GCN	To classify criminals from non-criminals
[39]	BN	To classify suspect users from non-suspect users
[103]	GBDT	To detect significant locations in users’ visiting patterns
[29,30]	RF	To classify geographical areas into two classes, high or low crime levels
[7]	RF	To predict population density in Portugal and France
[23]	RF	To classify urban areas in Tel Aviv
[104]	DBSCAN, GMM	The DBSCAN algorithm is used to cluster users’ trajectories into meaningful places, while GMM is used to identify users’ habits
[11]	SVM	To classify urban land use in Beijing into six classes, (a) residential, (b) business, (c) scenic, (d) open, (e) other, and (f) entertainment
[12]	K-means	To identify urban functional areas (UFAs) in Beijing
[105]	GAN	To create artificial maps of population density distributions
[106,107]	K-means	To classify city users based on their calling behaviors into different types of city geographic areas, including residents, visitors, and commuters
[108]	MLP	To predict the real estate price in Budapest, Hungary
[109]	RF, GBDT, SVM, Adaptive boosting	To reconstruct individual trajectories
[110]	MLP, CNN, LSTM	To predict crowd distributions of people in urban areas
[111]	NB, LR, RF, DT, KNN	To prompt or recommend the best mobile phone contract services based on customer communication behaviors
[112]	BP	To estimate individual exposure to particulate matter (PM2.5) air pollution
[113]	ADTree, FT, RF	To detect subscriber identity module box (SIMbox) fraud
[114]	LR, SVM-Linear, SVM-RBF, KNN, RF	To predict demographic features such as age and gender
[115]	SVM-Linear, Logistic regression	To predict demographic features such as age and gender
[116]	NB, SVM, DS, RF, RNN	To predict the next location of tourists
[117]	HC	To cluster human mobility patterns based on similar individual trajectories
[118]	GAN	To generate synthetic data of mobile phone data
[119]	KNN, RF, SVM-Linear, SVM-RBF+CNN, LSTM, SDAE	To construct a classifier that enables the recognition of fraudulent phone calls
[120]	RF, GBDT, SVM +CNN	To classify churner customers from non-churner customers
[121]	DT, RF, GBDT, XGBoost	To predict customer churn in Syriatel telecom company
[122]	MLP, SVM, Bayesian networks	To detect prepaid customer churn in mobile telecommunications companies
[123]	RF, DT, MLP, GBDT	To build predictive models that can classify customers into different categories of loyalty, such as very high value customers (greater loyalty), medium value customers (average loyalty), and others
[124]	LDA, SVM-RBF, XGBoost), RF, LR, NB, KNN, Bagged CART, CART, GBDT, C5.0	To predict customer demographic variables such as age and gender in Syriatel Telecom Company
[125]	K-means, DBSCAN	To detect fraudulent calls in telecommunications companies such as
[126]	GMM, ANN	To build a clustering-based classification model to classify cellular network traffic patterns into high-activity area, medium-activity area, low-activity area, etc.
[92,94,127]	K-means, GMM+CNN	To detect anomalous behavior through the identification of anomalous activities of mobile phone subscribers [92], to detect anomalies in a cellular network such as sleeping cells or unusual high call volume in a given region (traffic activity) [94]
[128]	FCM	To classify mobile subscribers based on extracting their calling features into three classes genuine, fraudulent, and suspicious
[129,130]	HC, k-means, FCM, SVM	To detect fraudulent behaviors in telecom companies such as detecting fraudulent calls
[10]	K-means, FCM, spectral clustering, consensus clustering	To cluster land use in Madrid
[131]	FKNN, MLP, C4.5, SVM GBDT, LR, RF, Adaptive boosting	To classify mobile customers into two classes, churners or non-churners
[132]	K-means	To cluster users according to their weekly mobility patterns into six different profiles