Predicting soil phosphorus and studying the effect of texture on the prediction accuracy using machine learning combined with near-infrared spectroscopy

QSAR modeling and molecular docking studies of 2-oxo-1, 2-dihydroquinoline-4- carboxylic acid derivatives as p-glycoprotein inhibitors for combating cancer multidrug resistance

“…This helps identify the key property of the molecule that affect its activity, suggest a structure or molecule with a specific activity as well as understand the interaction between functional groups in a molecule. Data modeling was performed, on centered and scaled data, with python core (v3.10) where several machine learning (ML) algorithms were used to develop predictive regression models including, decision tree (DT), k nearest neighbor (KNN), gradient boosting-based (GB) models, back propagation artificial neural network (BPNN) [ 36 , 37 , 38 ].…”

The synergic approach between machine learning, chemometrics, and NIR hyperspectral imagery for a real-time, reliable, and accurate prediction of mass loss in cement samples

“…Exploratory data analysis was then followed by the multivariate calibration, in other terms, ML models development that predicts a reference value Y (slow, time-consuming measurement and sometimes hard or impossible to obtain) from the corresponding recorded spectrum X (fast and easy measurement). To succeed in this stage, multiple regressors available in the literature including Partial Least Squares (PLSR) [ [20] , [21] , [22] , [23] ], Principal Components (PCR) [ 24 ], Support Vector Machine (SVMR) [ 20 , 25 , 26 ], Decision Tree (DTR) [ 27 ], K-Nearest Neighbors (KNNR) [ 28 , 29 ], eXtreme Gradient Boosting (XGBR) [ 30 ], Light Gradient Boosting Machine (LGBMR) [ 31 ], Categorical Boosting (CBR) [ 32 ], and MultiLayer Perceptron (MLPR) [ 33 ] with two hidden layers, were used for developing our predictive models. Due to the low number of samples, leave-triplicate-out Cross Validation (CV) [ 34 ] scores were adopted to assess the developed predictive models, overcome underfitting and overfitting problems as well as to drive models benchmarking study.…”

Incorporating soil knowledge into machine‐learning prediction of soil properties from soil spectra

“…Various methods have been used to select wavelengths in soil spectroscopy to reduce the model's complexity by removing the source of noise, and irrelevant variables in the spectra, thus improving the robustness of a calibration model (Zou et al, 2010). These methods include the successive projection algorithm (SPA) (Araújo et al, 2001), discriminant function analysis (Elliott et al, 2007), wavelet transform (Viscarra Rossel & Lark, 2009), continuous wavelet transform (Wang et al, 2016), genetic algorithm (Vohland et al, 2011), uninformative variable elimination (UVE) (Vohland & Emmerling, 2011), competitive adaptive reweighted sampling (Vohland et al, 2014), parallel factor analysis (Reda et al, 2020), backward variable elimination, iterative predictor weighting (Reda et al, 2019), mutual information algorithm (Zhang et al, 2019), and other approaches (Li et al, 2022). Yang et al (2012) achieved similar prediction performance with PLSR models using only three or four wavelengths produced by UVE coupled with SPA, compared with those based on all 2100 wavelengths.…”