Abeywickrama Bamunusinghe Kankanamge, Chathurika2025-06-262025-06-262025-06-26http://hdl.handle.net/10138/598202LASSO regression is an effective and regularized high-dimensional regression method which encourages sparsity by shrinking some regression coefficients to zero. This is a commonly used method for variable selection and estimation. However, classical inference techniques cannot be applied for LASSO estimates. This thesis evaluates the performance of different post-selection inference approaches which were used to obtain inference for LASSO after variable selection. A simulation study was conducted under two scenarios with different structures of non-null effects and data were generated using a block structured correlation matrix to model the realistic setting among predictors. Three post-selection inference methods used in this study are the sample splitting method, the conditional selective inference with theoretical penalization parameter, and the conditional selective inference with cross-validated penalization parameter. The performance of these methods was evaluated using metrics such as coverage probability, selective power, and selective type I error. Furthermore, this study examined the distribution of the confidence interval endpoints for each method. According to the results, the conditional selective inference method with cross-validated penalization parameter was identified as the most effective approach to obtain valid and reliable inference for LASSO estimates. The sample splitting method demonstrated the lowest performance among these methods, and the conditional selective inference method with theoretical penalization parameter performed better than the sample splitting method, but underperformed compared to the conditional selective inference method with cross-validated penalization parameter.engCC BY 4.0Post-selection inferenceLASSO regressionSample splittingConditional selective inferenceSimulation studyURN:NBN:fi:hulib-202506263068pro gradu -tutkielmaTilastotiedeStatisticsStatistikMatematiikan ja tilastotieteen maisteriohjelmaMaster's Programme in Mathematics and StatisticsMagisterprogrammet i matematik och statistik