Abstract | ||
---|---|---|
General theories describing the performance of artificial learners are of little help when a user is confronted with a selection of datasets and a given artificial classifier. The objective of this paper is to find out the best description of the learning curves produced by a Naive Bayes classification. The performance of Naive Bayes was measured on 121 datasets using k-fold cross-validation. Power, linear, logarithmic and exponential functions were fit to the data. The exponential function was a better descriptor of the error rate in 44 of 60 useful cases. Average mean squared error is significantly different at P=0,000 from power and linear and at P=0,001 from logarithmic function. The exponential function's rank is significantly different from the ranks of other models (P=0,000). The results can be used to forecast the future performance of the learner, or to check where on the learning curve the current measurement lies. |
Year | DOI | Venue |
---|---|---|
2014 | 10.1007/978-3-319-10933-6_20 | ADVANCES IN DATABASES AND INFORMATION SYSTEMS (ADBIS 2014) |
Keywords | Field | DocType |
Machine Learning,Power Law,Naive Bayes,Error rate,Learning curve | Data mining,Computer science,Mean squared error,Artificial intelligence,Logarithm,Classifier (linguistics),Exponential function,Pattern recognition,Naive Bayes classifier,Word error rate,Learning curve,Machine learning,Bayes classifier | Conference |
Volume | ISSN | Citations |
8716 | 0302-9743 | 0 |
PageRank | References | Authors |
0.34 | 16 | 3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Bostjan Brumen | 1 | 260 | 25.48 |
Ivan Rozman | 2 | 414 | 122.20 |
Ales Cernezel | 3 | 2 | 3.06 |