Improvement of the Fine tuning algorithm

Joseph Mietkiewicz, Anders Madsen

Research output: Contribution to journalArticlepeer-review

Abstract

Khalil El Hindi has developed a fine-tuning algorithm toimprove the classification accuracy of the Naive Bayes. His algorithm optimizes the conditional probability tables of the Naive Bayes after thetraining phase. The values of the probabilities of a variable are modified if it causes misclassification of a training instance. The algorithm out-performs in many cases the Naive Bayes. We analyze the performanceof the algorithm, discussed its issues, and compare it to a modified algorithm. The new algorithm simplifies the formula used in the fine-tuning algorithm and uses a more efficient scoring metric, the Brier score, tofine-tune the probabilities. The new algorithm shows an improvement in terms of classification accuracy on benchmark data sets compared to the Naive Bayes and fine tuned Naive Bayes.
Original languageEnglish
JournalUniversity of Antwerp
DOIs
Publication statusPublished - 2022

Keywords

  • fine-tuning algorithm
  • classification accuracy
  • Naive Bayes
  • conditional probability tables
  • training phase
  • misclassification
  • performance analysis
  • Brier score
  • benchmark data sets

Fingerprint

Dive into the research topics of 'Improvement of the Fine tuning algorithm'. Together they form a unique fingerprint.

Cite this