Skip to main content

Table 1 Machine learning models with hyperparameters

From: Publicly available machine learning models for identifying opioid misuse from the clinical notes of hospitalized patients

Model

Hyper-parameters

Logistic Regression-CUIs

C = 1, penalty = L1, class_weight = balanced

Logistic Regression-Words

C = 1, penalty = L1, class_weight = balanced

Convolutional Neural Network-CUIs

Filters = 1024, Filter Size = 1, Dropout = 0.5, Units = 1024, Learning Rate = 0.0001

Convolutional Neural Network-Words

Filters = 1024, Filter Size = 3, Dropout = 0.25, Units = 128, Learning Rate = 0.0001

Convolutional Neural Network-Character

Filters = 1024, Filter Size = 11, Dropout = 0.25, Units = 1024, Learning Rate = 0.0001

Deep Averaging Network-CUIs

Dropout = 0.25, Units in layer 1 = 2048, Units in layer 2 = 512, Learning Rate = 0.001

Deep Averaging Network-Words

Dropout = 0.75, Units = 128, Learning Rate = 0.001

Max Pooling Network-CUIs

Dropout = 0.5, Units = 128, Learning Rate = 0.001

Max Pooling Network-Words

Dropout = 0.5, Units = 128, Learning Rate = 0.001

Deep Averaging + Max Pooling Network-CUIs

Dropout = 0.5, Units = 1024, Learning Rate = 0.001

Deep Averaging + Max Pooling Network-Words

Dropout = 0.25, Units = 512, Learning Rate = 0.001

  1. Logistic regression’s C value is inverse of regularization strength, and penalty term that penalizes the loss function using different regularization techniques. Optimizer Adam is selected for all the neural networks. Units are the number of neurons in the dense layer of the neural network