Feature Selection: Do we need it?

This evening, it was my turn to give a presentation. And my topic was Feature Selection: Do we need it?

I had decided the paper only before 6 days, just after Shiqi Zhao's turn. I chose the papers about feature selection. I had experienced on this topic. On my anaphora resolution research, I had used some features. But for ACE, I found out that 28 features were better than 31 features. However, 29 features were better than 28 features. That time, I had not any idea about feature selection. So finally, I used 29 features. Maybe if I used some feature selection algorithms, I would obtain some better results on ACE.

After a short survey about feature selection research, I chose a paper which had been published on the Journal of Machine Learning Research. The paper was Lei Yu and Huan Liu. "Efficient Feature Selection via Analysis of Relevance and Redundancy". Journal of Machine Learning Research, 5(Oct), 1205-1224, 2004. It was of 20 pages. So it was a challenge for me!

Although I had practiced on my slides two times, I fell little tense at beginning. Maybe after one minute, I was in the nice state.

The outline of my presentation was as figure 1.


Figure 1. Outline of my presentation on Feature Selection: Do we need it?

My presentation slides' link was in my Homepage's My research related materials.

I had spent 55 minutes for my presentation. And finally, I made a short demo about hot to using the software.

I believed I had tried my best for the presentation. It was not only a presentation about feature selection, but also a beginning of research on feature selection.