Seminar: Understanding Dataset Learning Performance

Symbolbild zum Artikel. Der Link öffnet das Bild in einer großen Anzeige.


It is considered impossible to know the maximum possible learning performance of a machine learning (ML) algorithm on a given dataset before running the algorithm. The performance of an ML algorithm depends on many factors, such as the quality and size of the data, the complexity of the underlying patterns, the choice of algorithm, and the choice of hyperparameters. However, the “Bayes error rate” or the “Bayes optimal error” theorem estimates any algorithm’s minimum achievable error rate on a given dataset. The Bayes error rate is the lowest possible error rate that can be achieved by any classifier, assuming the optimal classifier that perfectly separates the classes is available.


Develop an algorithm that can efficiently estimate the minimum error rate level in a target
class for any given dataset with categorical variables.

Learning Objectives

  • Explore interactions between rule mining strategies and neural network architectures in the context of error estimation
  • Apply different concepts of error and performance metrics in machine learning
  • Create an efficient algorithm to compute any dataset’s minimum error rate


Project typeSeminar
ECTSFor seminars: 2.5, 5, 7.5, default: 5
PeriodSummer term 2024
Presence timeVirtual seminar, working from remote
Useful knowledgePython
Work distribution25% algorithm design, 50% programming, 25% running experiments
Med. Eng. designationAdvanced Context Recognition (ACR)
StudOn link
First meeting
RegistrationVia StudOn, obligatory after introduction.


Up-to-date literature recommendations are provided during the lectures.


  • Final project presentation, demonstrator and final report.


Dr. Luis I. Lopera G.

  • Job title: Researcher
  • Address:
    Henkestraße 91, Haus 7, 1. OG
    91052 Erlangen
  • Phone number: +49 9131 85-23605
  • Email:

Friedrich-Alexander-Universität Erlangen-Nürnberg