Automatic selection of classification algorithms usingmeta-features and SMOTE  

Oleh:

Otsuka, Atsushi ; Nakamura, Munehiro ; Nambo, Hidetaka ; Kimura, Haruhiko ; Todo, Yuki

Jenis: Article from Proceeding
Dalam koleksi: The 14th Asia Pacific Industrial Engineering and Management Systems Conference (APIEMS), 3-6 December 2013 Cebu, Philippines, page 1-6.
Topik: datamaining ; SMOTE ; classifier
Fulltext: 1192_Oyabu.pdf (643.78KB)

Isi artikel

Classification of a problem plays an important role in the real world. However, currently it is difficult to find the best classification algorithm for a classification problem in a large variety of classification algorithms. . This paper presents a recommender system of five standard classification algorithms, namely Support Vector Machine, Random Forest, Neural Network, k nearest neighbor algorithm, and Naive Bayes. With this system, it is expected that users specifically for non-experts would easily find good classification algorithms and shorten their searching time. The proposed system first extracts basic meta-features from various datasets. Using the meta-features the proposed method estimates the best classification algorithm for a dataset. We have prepared 73 real-world datasets for evaluating the proposed system and conducted leave-one-out cross validation: choosing 72 datasets as learning data and one dataset as test data. However, since we found that the learning data was small to find the best classification algorithm, we applied a over-sampling method called SMOTE to the learning data. Evaluation experiments for the 73 datasets have shown that the proposed system chooses the best classification algorithm at 56.16% with SMOTE and 10.96% without SMOTE in average.

Opini Anda

Klik untuk menuliskan opini Anda tentang koleksi ini!

Kembali