Laplace Kernel Adaptive Tuning Optimized Transfer Learning for Cardiovascular Disease Prediction

Cardiovascular diseases (CVDs) are common diseases
of the heart and blood vessels, accounting for devastating health
outcome and mortality worldwide. The risk factors are
hypertension, hyperlipidaemia, tobacco smoking and diabetes.
For health benefits to take place, diagnosis needs to be done early
for early intervention. While machine learning and deep learning
have demonstrated their applications in CVD prediction, their
CVD prediction accuracy and their need for minimal error and
short computation time have been problematic. For this purpose,
a Laplace Kernel Stochastic Tuned Optimized Transfer Learning
(LKSTOTL) is proposed. The framework consists of the subsections
of data acquisition, pre-processing, feature selection,
classification and fine tuning. Patient data are collected,
preprocessed to remove missing values and outliers and for
dimensionality reduction, significant feature selection using
Laplace kernelized stochastic neighbor embedding method is
carried out. The pre-trained model is taken from a deep belief
network (DBN), and the transfer learning is employed to enhance
learning efficiency of the features. The training and test errors
are minimized with the spiral search optimization algorithm.
Experimental evaluation indicates that LKSTOTL is superior
over the conventional models with 8% accuracy improvement,
5% precision improvement, 4% recall improvement, 4% F1-
score improvement, 10% specificity improvement, and 77%
error rate and 18% prediction time improvement, showing its
application for reliable and efficient cardiovascular disease
prediction.