Learning rules from numerical data by combining geometric and graph-theoretic approach
✍ Scribed by Sukhamay Kundu; Jianhua Chen
- Publisher
- Elsevier Science
- Year
- 2000
- Tongue
- English
- Weight
- 317 KB
- Volume
- 33
- Category
- Article
- ISSN
- 0921-8890
No coin nor oath required. For personal study only.
✦ Synopsis
We present a new method for learning rules from numerical data by using a combination of geometric and graph-theoretic methods. Three different graphs are defined to capture the geometric properties of the data-set. The graphs G(P ) and G(N ) capture the intra-class properties of the set of positive points P and the set of negative points N , respectively, and the graph G(P , N) captures the inter-class properties between P and N . We derive rules from these graphs by means of graph partitioning. Our method tends to give fewer rules than that obtained by decision-tree based methods. The complexity of our algorithm is O(M 3 ), where M = |P | + |N |.