AINDRILA GARAI

– Objective: To analyze extreme value data we have worked on Generalized Pareto Distribution (GPD).

– Compared the Absolute Bias and rMSE for the ZJ-method, Maximum Likelihood Estimator (MLE), Elementary Percentile Method (EPM), M-estimator, and a weighted version of M-estimator to estimate the shape and the scale parameters of GPD.

– Interval estimation using Bootstrapped-t method.

– Trying to develop a new method to estimate the parameters more accurately for the whole range of parameters.

Objective: Predicting true class labels using a surface-to-volume ratio to tackle the curse of imbalanced data.

– Developed two novel methods Bagged Regularized Decision Tree (BaRDT) and Boosted Regularized Decision Tree (BoRDT), base learners as SVRTrees that use weighted probabilities and SVR as splitting criteria. – Calculate the Vapnik-Chervonenkis (VC) dimension of SVRTree as a requirement of consistency proof of BoRDT.

– Compared isometrics for the different distance functions that are implemented in Decision Trees to depict relative trade-offs between true positive and false positive.

– We tested our methods with the baseline imbalanced classification models on 23 imbalanced datasets and 27 balanced datasets based on AUC-ROC and F-measure.

– We have extended these two ideas for Multi-category Classification. – We have implemented Conformal Prediction for classification with Model Agnostic Prediction Interval Estimator (MAPIE) on these setups and got a significant result.

– The analysis suggests that BoRDT & BaRDT generally outperform SVR Tree and other models

– Prepared notes on the primary concepts of Machine Learning by reading the topics of the books ISLR and ESLR with accompanying code examples.