@inproceedings {pub3603pub3693,
	title = {Enhancing Very Fast Decision Trees Using Local Split-Time Prediction},
	author = {Viktor Losing AND Heiko Wersing AND Barbara Hammer},
	year = {2018},
	abstract = {An increasing number of industrial areas recognize the opportunities of Big Data and try to utilize it for their business. However, they also face the related challenges as the excessively growing information flood, ever more requiring highly efficient algorithms which enable real-time processing and reduce the burden of data storage and maintenance. Decision trees are extremely fast, highly accurate and easy to use in practice. Utilized within an ensemble they are one of the most powerful machine learning methods. The Very Fast Decision Tree is the state-of-the-art incremental decision tree induction algorithm, capable of learning from massive data streams. In this paper, we increase its efficiency even further by replacing its splitting scheme, which periodically tests each n min examples whether a leaf can be split. Instead, we utilize the local statistics to predict the split-time, thus, avoiding unnecessary split-attempts, usually dominating the computational cost. We propose two different approaches each targeting different scenarios. We extensively evaluate our methods using commonly applied benchmarks consisting of i.i.d. streams as well as those incorporating concept drift. The experiments confirm a significantly reduced run-time without a loss in classification performance.},
	publisher = {IEEE},
	booktitle = {ICDM 2018}
}