On Benchmark Experiments and Visualization Methods for the Evaluation and Interpretation of Machine Learning Models

MCML Authors

Giuseppe Casalicchio

Dr.

→ Group Bernd Bischl
 Statistical Learning and Data Science

Abstract

This cumulative dissertation consists of five articles divided into three parts. The first part extends the mlr package in R to implement and benchmark multilabel classification methods. The second part focuses on simplifying benchmark experiments with OpenML.org, introducing the OpenML R package and the OpenML100 benchmarking suite for standardized dataset and result management. The third part addresses model evaluation and interpretability, proposing the residual-based predictiveness (RBP) curve to improve upon the predictiveness curve and introducing new visualization tools, including the Shapley feature importance (SFIMP) measure for model interpretation. (Shortened.)

phdthesis

---

Dissertation

LMU München. Mar. 2019

---

Authors

G. Casalicchio

---

Links

DOI

---

Research Area

 A1 | Statistical Foundations & Explainability