Selecting the most important self-assessed features for predicting conversion to Mild Cognitive Impairment with Random Forest and Permutation-based methods

Gómez-Ramírez, Jaime; Ávila-Villanueva, Marina; Fernández-Blázquez, Miguel A.

doi:10.1101/785519

Cited by 17 publications

(22 citation statements)

References 29 publications

(15 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…It performs better for large, high-dimensional data sets and is more robust to noise and feature selection [34,[61][62][63]65]. Additionally, the prediction ability of Random Forest is resistant to the multicollinearity of the driven variables [66][67][68], so all the all the variables in Equation (2) were used to predict the spatially continuous TanSat SIF. The main parameters used in RF are the numbers of input prediction variables and decision trees.…”

Section: Random Forest Approach For Sif Modelingmentioning

confidence: 99%

Generation of a Global Spatially Continuous TanSat Solar-Induced Chlorophyll Fluorescence Product by Considering the Impact of the Solar Radiation Intensity

Liu

Chen

et al. 2020

Remote Sensing

View full text Add to dashboard Cite

Solar-induced chlorophyll fluorescence (SIF) provides a new and direct way of monitoring photosynthetic activity. However, current SIF products are limited by low spatial resolution or sparse sampling. In this paper, we present a data-driven method of generating a global, spatially continuous TanSat SIF product. Firstly, the key explanatory variables for modelling canopy SIF were investigated using in-situ and satellite observations. According to theoretical and experimental analysis, the solar radiation intensity was found to be a dominant driving environmental variable for the SIF yield at both the canopy and global scales; this has, however, been neglected in previous research. The cosine value of the solar zenith angle at noon (cos (SZA0)), a proxy for solar radiation intensity, was found to be a dominant abiotic factor for the SIF yield. Next, a Random Forest (RF) approach was employed for SIF prediction based on Moderate Resolution Imaging Spectroradiometer (MODIS) visible-to-NIR reflectance data, the normalized difference vegetation (NDVI), cos (SZA0), and air temperature. The machine learning model performed well at predicting SIF, giving R2 values of 0.73, an RMSE of 0.30 mW m−2 nm−1 sr−1 and a bias of 0.22 mW m−2 nm−1 sr−1 for 2018. If cos (SZA0) was not included, the accuracy of the RF model decreased: the R2 value was then 0.65, the RMSE 0.34 mW m−2 nm−1 sr−1 and an bias of 0.26 mW m−2 nm−1 sr−1, further verifying the importance of cos (SZA0). Finally, the globally continuous TanSat SIF product was developed and compared to the TROPOspheric Monitoring Instrument (TROPOMI) SIF data. The results showed that the globally continuous TanSat SIF product agreed well with the TROPOMI SIF data, with an R2 value of 0.73. Thus, this paper presents an improved approach to modelling satellite SIF that has a better accuracy, and the study also generated a global, spatially continuous TanSat SIF product with a spatial resolution of 0.05°.

show abstract

Section: Random Forest Approach For Sif Modelingmentioning

confidence: 99%

Generation of a Global Spatially Continuous TanSat Solar-Induced Chlorophyll Fluorescence Product by Considering the Impact of the Solar Radiation Intensity

Liu

Chen

et al. 2020

Remote Sensing

View full text Add to dashboard Cite

show abstract

“…Although there is a study using random forest and permutation-based methods to select important variables for predicting conversion to MCI ( Gómez-Ramírez et al, 2019 ), to the best of our knowledge, this is the first study investigating the predictive rather than associative value of the odors in the OI test for incident dementia in the elderly. There are several strengths in our study.…”

Section: Discussionmentioning

confidence: 99%

“…To mask the information on a variable during validation, instead of removing the variable from the data set, the PI method replaces it with random noise by shuffling the values of the variable, i.e., using values from other participants ( Breiman, 2001 ; Fisher et al, 2019 ). The relative importance of a variable was calculated as the accuracy decrease of the variable relative to the range of the accuracy decreases of all the variables ( Gómez-Ramírez et al, 2019 ).…”

Section: Methodsmentioning

confidence: 99%

Predictive Value of Odor Identification for Incident Dementia: The Shanghai Aging Study

Ding

Xiao

Liang

et al. 2020

Front. Aging Neurosci.

View full text Add to dashboard Cite

Objective: This study aimed to evaluate the value of odors in the olfactory identification (OI) test and other known risk factors for predicting incident dementia in the prospective Shanghai Aging Study. Methods: At baseline, OI was assessed using the Sniffin' Sticks Screening Test 12, which contains 12 different odors. Cognition assessment and consensus diagnosis were conducted at both baseline and follow-up to identify incident dementia. Four different multivariable logistic regression (MLR) models were used for predicting incident dementia. In the no-odor model, only demographics, lifestyle, and medical history variables were included. In the single-odor model, we further added one single odor to the first model. In the full model, all 12 odors were included. In the stepwise model, the variables were selected using a bidirectional stepwise selection method. The predictive abilities of these models were evaluated by the area under the receiver operating characteristic curve (AUC). The permutation importance method was used to evaluate the relative importance of different odors and other known risk factors. Results: Seventy-five (8%) incident dementia cases were diagnosed during 4.9 years of follow-up among 947 participants. The full and the stepwise MLR model (AUC = 0.916 and 0.914, respectively) have better predictive abilities compared with those of the no-or single-odor models. The five most important variables are Mini-Mental State Examination (MMSE) score, age, peppermint detection, coronary artery disease, and height in the full model, and MMSE, age, peppermint detection, stroke, and education in the stepwise model. The combination of only the top five variables in the stepwise model (AUC = 0.901 and sensitivity = 0.880) has as a good a predictive ability as other models. Conclusion: The ability to smell peppermint might be one of the useful indicators for predicting dementia. Combining peppermint detection with MMSE, age, education, and history of stroke may have sensitive and robust predictive value for dementia in older adults.

show abstract

“…Variables’ importance was evaluated using both the permutation importance (PI) by the multivariate logistic regression (LR) model and Gini importance (GI) by the random forest (RF) model (27-29) (eMethods 1). We used the K-fold cross-validation method to calculate the predictive ability indices (30) (eMethods 2), including accuracy, sensitivity, specificity, and area under the receiver operating characteristic (ROC) curve (31).…”

Section: Methodsmentioning

confidence: 99%

Poor odor identification predicts mortality risk in older adults without neurodegenerative diseases: the Shanghai Aging Study

Xiao

Zhao

Liang

et al. 2020

Preprint

View full text Add to dashboard Cite

BackgroundIrrespective of neurodegeneration, the decline of olfactory function might also reflect a wider range of pathological conditions contributing to mortality. However, the potential explanations and their predictive values have rarely been reported.MethodsA total of 1,433 older adults aged ≥ 60 years without neurodegenerative disease were administered a follow-up of 8.6 years on average. Sniffin’ Sticks Screening Test was used to assess the olfactory identification at baseline. Survival status during the follow-up was obtained from the local Center for Disease Control and Prevention. Bidirectional stepwise Cox proportional hazards regression was used to identify variables associated with mortality. Two predictive models were constructed by statistical learning methods.ResultsAll-cause mortality rate was 1.1/100 person-years during the follow-up. Sex (HR = 0.65, 95%CI 0.46 - 0.93), age (HR = 1.12, 95%CI 1.10 - 1.15), chronic kidney disease (HR = 1.88, 95%CI 1.09 - 3.25), low density lipoprotein (HR = 0.81, 95%CI 0.67 - 0.99), anemia (HR = 2.74, 95%CI 1.19 - 6.30), and fail to identify coffee odor (HR = 1.96, 95%CI 1.19 - 3.23) were significantly associated with all-cause mortality. The logistic regression and the random forest predictive models showed similar predictive accuracy (0.91 and 0.90) and area under the receiver operating characteristic curve (0.77 and 0.75).ConclusionsThe association between poor olfactory and long-term mortality was verified among Chinese older population. Certain odors identification ability may contribute to the prediction of long-term mortality along with other important risk factors.

show abstract

Selecting the most important self-assessed features for predicting conversion to Mild Cognitive Impairment with Random Forest and Permutation-based methods

Cited by 17 publications

References 29 publications

Generation of a Global Spatially Continuous TanSat Solar-Induced Chlorophyll Fluorescence Product by Considering the Impact of the Solar Radiation Intensity

Generation of a Global Spatially Continuous TanSat Solar-Induced Chlorophyll Fluorescence Product by Considering the Impact of the Solar Radiation Intensity

Predictive Value of Odor Identification for Incident Dementia: The Shanghai Aging Study

Poor odor identification predicts mortality risk in older adults without neurodegenerative diseases: the Shanghai Aging Study

Contact Info

Product

Resources

About