Login Register



Auto-login on future visits

Forgot your password?

Newsletter May-June 2010

How to optimize multi-class classifiers in imbalanced problems?

Optimization of multi-class classifiers in imbalanced problems.

Tutorial Multi class ROC

When one of the classes is much larger than others, small classes are often heavily misclassified. In this tutorial example, we show how to tune multi-class classifiers to take into account such heavy class imbalance. The key tool is multi-class ROC analysis. It tunes the weighting factors applied to real-value outputs of statistical models so that a more favorable trade-off between class errors is achieved. The example illustrates the use of PRSD Studio interactive ROC analysis tools allowing you to explore classifier performance at different operating points.

See the video in our blog

Student and academic versions available now!

Announcements

PRSD Studio is now available for academic researchers and students. Why to adopt PRSD Studio in your research?

  • Save time and focus on your core research with robust interactive tools
  • Easily compare with state-of-the-art pattern recognition algorithms
  • Quickly create engaging demos
  • Benefit form powerful and unique evaluation and optimization tools (ROC analysis)
Different academic licenses are available including student and classroom options. See more.

Automatic 3D particle shape classification

DeBeers product

Customer story

DebTech, a company within the DeBeers mining group, developed their R-Sputnik particle characterization system using PRSD Studio. The system captures silhouette images of particles and generates 3D models of the particles.

Using PRSD Studio 2.1 we have implemented a classifier that takes in various features extracted from these 3D particle models and classifies them into a number of shape categories that are of relevance to us. The system also allows us to take simultaneous reflectance image of each particle. Relevant data from these images is fed into another classifier that enables us to obtain a colour classification for each particle. -- Garry Morrison, DebTech.

More application examples

Visualizing live class distributions in scatter plots

Visualization of class distributions in scatter plots.

Feature Highlight

PRSD Studio 2.2.1 enhances the interactive scatter plots with live class distributions. Separate plots rendering class distributions are shown for each of the features (axes) in the scatter plot. The class distribution plots highlight the nature of class overlap. This is especially beneficial in large data sets where scatter plot alone is not sufficiently informative. The class distribution plots are updated live when changing features, focusing on sample subsets or painting new class labels.

See the video in our blog