Skip to main content

Table 5 The best parameters for the tested models. “States” and “Threshold” are the hidden state number and convergence threshold of the EM algorithm of the HMM. “n-gram” is the gram size in the preprocessing for the NB and SVM models, “IDF” is if to include inverse document frequency (see Sect. 4.2.2) in the preprocessing for the NB and SVM models, and “C” is the penalty parameter of the error term for the SVM models. “Embed” is the hidden node number in the embedded layer (see Sect. 4.1), “DOi” is the ith fraction of the input units to drop in the dropout layer, “Filter size” is the size of the filter in the convolution layer, “# filters” is the number of filters in the convolution layer, “# Pooling” is the number of neurons to pool in the pooling layer, “Neuron” is the hidden node number in the CNN model, and “Memory” is the memory unit number of the LSTM

From: Identifying and predicting social lifestyles in people’s trajectories by neural networks

  States Threshold       
HMM 30 0.01       
  n-gram IDF       
NB 14 TRUE       
  n-gram IDF C      
LSVM 14 TRUE 10      
RSVM 2 TRUE 10      
PSVM 2 FALSE 0.2      
SVC 14 TRUE 10      
  Embed DO1 Filter size # filters DO2 Neuron   
CNN 1300 0.2 1000 7 0.2 1150   
  Embed DO1 Memory DO2     
RNN 650 0.4 500 0     
LSTM 550 0.8 1050 0.2     
BLSTM 550 0.8 1250 0     
  Embed DO1 Filter size # filters # Pooling DO2 Memory DO3
CLSTM 1100 0.6 550 7 2 0.6 1250 0.6
CBLSTM 600 0.2 950 7 2 0.6 1250 0.4