Integrating Feature Engineering with Deep Learning to Conduct Diagnostic and Predictive Analytics for Turbofan Engines

Table 4

Adjustable hyperparameters in machine learning and deep learning.

Model	Adjustable hyperparameters	Range
MARS	Degree of nonlinearity	1–3
	Penalty	1–30
RF	Number of trees	500–1000
	Maximal depths	4–36
	Maximal features	0.5–0.9
	Minimal samples split	4–96
	Minimal samples leaf	4–72
XGB	Number of trees	500–1000
	Maximal depth	4–36
	Minimum children weight	1–5
	Gamma	0–0.3
	Subsample	0.6–1
	Column sample by tree	0.6–1
	Learning rate	0.0025–0.005
SVM	Kernel	RBF, sigmoid
	Cost penalty	1–50
	Gamma	0.01–1
	Epsilon	0.1–1
DNN, RNN, LSTM, GRU, CNN ( denotes additional parameters only required for CNN)	Hidden layers	1–4
	Neurons	10–200
	Dropout	0.1–0.5
	Activation	ReLu, tanh, softplus
	Optimizer	Adam, RMSprop
	Learning rate	0.0001, 0.001, 0.01
	Epochs	500
	Batch size	32, 64, 128, 256, 512
	Convolution layers	1–2
	Pooling layers	1–2
	Filters size	16, 32, 64, 128
	Fully connected layers	1–4