Classification-using-KMeans-and-SimpleMKL

Very large scale classification based on K-Means clustering & Multi-Kernel SVM(SimpleMKL)

Here, we are going to implement the method proposed in this article, "Very large scale classification based on K-Means clustering & Multi-Kernel SVM(SimpleMKL)" at ACM Digital Library.

Modules:

The code has below modules:

• KMeans Clustering
  • Select nearest & furthest points of each cluster
• Duplicate Removal
  • Remove all duplicate data
• Outlier Detection
  • Remove the last ROT-data based on their outlier score
  • Method proposed in this article, "Robust, Scalable Anomaly Detection for Large Collections of Images".
• Human Labeling
  • Do labeling for the new representative dataset
• SimpleMKL
  • Multi Kernel SVM
  • Method proposed in this article, "Simplemkl".

Datasets:

The method is run on two diffrent types of datasets, large scale & very large scale satasets.

The large scale datasets are:

• Breast-W (683*10)
• Messidor (1151*20)
• Car (1728*6)
  • uacc vs other
• Spambase (4601*57)

The very large scale datasets are:

• Coil2000 (9’822*85)
• Bank Marketing (45’211*17)
• Skin Segmentation (245’057*4)
• Covertype (581’012*54)
  • Aspen vs other

Results:

Results can be seen at the end of presentation file uploaded in this repository.