Use of random forest in FTIR analysis of LDL cholesterol and tri‐glycerides for hyperlipidemia

Abstract

A quantitative determination method for the diagnosis of hyperlipidemia was developed using Fourier transform infrared (FTIR) spectroscopy. Random forest (RF) was demonstrated as a potential multivariate algorithm for the FTIR analysis of low-density lipoprotein cholesterol (LDL-C) and tri-glycerides (TG) in human serum samples. The informative wavebands for LDL-C and TG were selected based on the Gini importance. The selected wavebands were mainly within the fingerprint region. The RF modeling results were better than those derived using PLS in validation process, because the chance for over-fitting was possibly eliminated in RF algorithm. ARF also demonstrated favorable results in the test process. The prospective model exhibited a higher than 90% true prediction in negative/positive properties for male and female samples. These clinical statistical results indicated the optimization of RF algorithm performed accurately in the FTIR determination of LDL-C and TG. RF is evaluated as a promising tool for diagnosing and controlling hyperlipidemia in populations. The parameter optimization methodology is useful in the improving model accuracy using FTIR spectroscopic technology.

Spatiotemporal variability of sunshine duration and influential climatic factors in mainland China during 1959–2017

“…The Gini index, a general indicator of feature relevance, visualizes the outcome of implicit feature selection (Menze et al, 2009). Based on the Gini index, one split of a tree node is created according to the balance between its own variables and the variables of its two descendent nodes (Chen et al, 2015), and the heterogeneity and relevancy of variables can be measured with CART and RF. The MDG denotes the average decrease of the Gini index in all trees for each feature, and is considered to be a robust and innovative index to exert variable selection because of its simplicity, as well as its ability to rapidly compute classification and regression (Han et al, 2016;Nembrini et al, 2018).…”

Raman Based Molecular Imaging and Analytics: A Magic Bullet for Biomedical Applications!?

“…With various kernels the SVM can be converted to a nonlinear classification and regression model. − Another example of powerful classification and regression algorithms are random forests (RF), which is an ensemble based method. RFs are based on the idea, which is related to automatic generated decision trees: − A predefined number of random decision trees are constructed and for prediction every tree is allowed to predict. The output of the whole random forest is generated by a voting procedure at the end.…”

PRFFECT: A versatile tool for spectroscopists

“…Tester) we provide a robust methodology to make it easier to test different pre- The Random Forest (RF) machine learning algorithm is widely used in many different fields of research, including cheminformatics, [10,11] bioinformatics, [12], and ecology, [13]. Within the field of biomedical spectroscopy, RF has been used in the annotation of lung cancer subtypes [14] and in the diagnosis of nonsmall cell lung carcinoma, [15] urinary bladder cancer, [16], hyperlipidemia [17], and brain tumours [7]. RF has proven to be a robust and accurate technique for developing spectral diagnostic models, giving excellent classification results without over-fitting.…”