1 / 13

Machine learning with Weka for Kung fu - training purposes

Machine learning with Weka for Kung fu - training purposes. Victoria Værnø victoriavarno@gmail.com v aernoe.wordpress.com. Background. Motion capture and machine learning workbenches are accessible to the public .

arawn
Download Presentation

Machine learning with Weka for Kung fu - training purposes

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Machine learning with Weka for Kung fu-training purposes Victoria Værnø victoriavarno@gmail.com vaernoe.wordpress.com

  2. Background Motion capture and machinelearningworkbenchesareaccessible to thepublic. Links betweencheap hardware (eg. web-cams and Kinect) & opensourcemachinelearningsoftware is still hard to find for motion capture.

  3. Research focus • Build a model which is good enough to consider for a beta application for amateur Kung-Fu training purposes. • For separately created unseen data, high accuracy and high “bad”-class label precision • What challenges follow a relatively limited training set and what basic machine learning techniques can reduce these?

  4. Method Meta- Learner Clustering Test Result- and Data Analysis

  5. Data Data attributes: 18 joints * 3 dimensions * 6 frames per movie + 1 classlable = 325 attributes .bvh .arff

  6. Iteration 1 Generate MLP model on the training data. Accuracy 97% Precision • of”Bad” 1 Good! Too good?

  7. Iteration 1 Test the model with unseen dataset. Accuracy 85% Precision • of”Bad” 1 Not so great. Unbalanced dataset?

  8. Unbalanced Data Attributevalueswhich in realitydefineclass 1 Class 1 Class 2 Attributevalueswhichveryoftenarise in class 1, butdoesnotdefineclass 1. Attributevalueswhich in realitydefineclass 2 This part is muchbigger in the test set and real life.

  9. Iteration 2 Clustering K-means K=3

  10. Trying to compensate for unbalanced data • Meta-classifier AdaBoostM1 boosting algorithm • Collecting more data – new people. • Other suggestions, please tell or email me! (victoriavarno@gmail.com)

  11. Iteration 3 Generate models and test on Data set 2. Just MLP + Data set 1 AdaBoostM1 84% AdaBoostM1 + Data set 1 80% Just MLP + Data set 1 + Data set 3 + 89% AdaBoostM1 + Data set 1 + Data set 3 86%

  12. Conclusion • Promisingresultsfor furtherinvestigation. • Cross-validation over 90% • ”Bad” classlabelalwaysprecision = 1 • Adding just one more person -> unseen data 89% • Classifyingunseenpeople’s kicks remainunexplored. • Data collection: Hard for one person to create balanced motion data. • Modeling: Boosting strategy to combat unbalanced motion capture data works in some cases, but adding a different person’s motion is far more efficient. ”Spend time gathering more data rather than tuning a particular method” Nilsson N.J Lessons Learned

  13. Thank You! Q?

More Related