Create a free account to run LASSO
Medhesap membership is completely free and takes seconds. You will return to this page after signing in.
No credit card · Your data is never shared
Reklamsız kullanım deneyimi için lütfen üye olunuz veya giriş yapınız.
LASSO (Least Absolute Shrinkage and Selection Operator) adds a penalty on the sum of the absolute regression coefficients. As the penalty grows, coefficients of weakly associated variables shrink to exactly zero and those variables leave the model, so the method performs shrinkage and feature selection at the same time. For binary outcomes it is applied as L1-penalized logistic regression (Tibshirani, 1996).
When there are many candidate predictors relative to the number of events (events per variable, EPV, below 10) or the predictors are correlated (e.g. CRP and CRP/albumin ratio), LASSO gives a more stable variable list that is less prone to overfitting than picking variables with p < 0.05 in univariable tests.
The penalty λ is tuned with stratified k-fold cross-validation. λ.min gives the lowest prediction error; λ.1se is the most parsimonious model within one standard error of it. Mark the variable of your primary hypothesis as “forced”: it is left unpenalized, stays in the model at every λ, and the other predictors are selected after adjusting for it.
The tool lists the selected predictors, standardized coefficients, the order in which variables enter the model and their bootstrap selection frequency. Copy the selected variables to SPSS in one line to build your multivariable logistic regression, and use the ready-made methods paragraph in English or Turkish, filled in with your settings.
No. Data is sent to the server over an encrypted connection and discarded from memory as soon as the analysis finishes. Only anonymous usage counts (e.g. number of patients analysed) are kept. Still, do not paste identifying patient information.
The LASSO penalty acts on the size of the coefficients, so every predictor is converted to a z-score to penalize variables on different scales (e.g. age and CRP) equally. Within cross-validation the scaling uses only the training fold. Reported coefficients are on the standardized scale.
It means that, after adjusting for the forced variable, no other predictor reliably improves prediction. Try the λ.min rule, or use the “First N variables” option to take a fixed number of predictors in the order they enter the model. Keep events per variable (EPV) above 10.
Variables with more missing values than your threshold (default 30%) are excluded; patients with missing values in the remaining variables are left out (complete-case analysis). Forced variables are never excluded by the threshold. If you need multiple imputation (MICE), complete the data in SPSS or R beforehand.
The data are resampled hundreds of times keeping the outcome ratio, and LASSO is refitted at the same λ in each resample. A variable's selection percentage shows how consistently it is chosen; variables below 60% are considered borderline.
A single analysis supports up to 5,000 rows and 150 model terms (each level of a categorical variable counts as one term). For large data the number of folds or λ values may be reduced automatically to stay within the time limit; a note is shown with the results.
References: Tibshirani R. Regression shrinkage and selection via the lasso. J R Stat Soc B. 1996;58(1):267-88 · Friedman J, Hastie T, Tibshirani R. Regularization paths for generalized linear models via coordinate descent. J Stat Softw. 2010;33(1):1-22 · Meinshausen N, Bühlmann P. Stability selection. J R Stat Soc B. 2010;72(4):417-73.
Hesap Araçları
AI Asistan