What are your inputs to your model?

czar_iv said:

---

What are your inputs to your model?

---

RPI, SOS, Regular season win %, last 10 games win %, seed, and conference. Would have loved to add a bunch more but I did this in about 3 days in my free time. I'll start much earlier next year.

The data used is from 2000 - 2018 to train/test it.

I was in the same boat. I've never done Kaggle competitions before so I thought this would be fun. Put together a rough model over last weekend.

Don't remember what all I wound up using off the top of my head, but it was something like:

• Win %
• Seed
• Offensive 3PT Rate
• Defensive 3PT Rate
• Rebounding Differential
• ORating
• DRating
• Pace

I wanted to do a lot more and maybe next year I'll take the full play by play data and see what kind of interesting things I can find through that. The test accuracy on the model that includes the above was also in the low 70% range, so it's interesting that there's not much difference between our two methods.

One thing that surprised me was that I one hot encoded the conferences and it made almost no difference in the test accuracy.

If you'd like, PM me and I ca share my final transformed data and you can use it in your model. Would be interesting to see if any better accuracy came out of it.

This is pretty cool - love to see some metrics.

boboguitar said:

---

One thing that surprised me was that I one hot encoded the conferences and it made almost no difference in the test accuracy.

If you'd like, PM me and I ca share my final transformed data and you can use it in your model. Would be interesting to see if any better accuracy came out of it.

---

I would assume the impact conference makes is mostly accounted for with SOS.

So how is this workin' for ya ???

EXCELL said:

---

So how is this workin' for ya ???

---

78.7% correct as of now. A little over test accuracy. It was 72% at the end of yesterday.

There's a lot more inputs I could have fed it given more time. It just takes a while to build/transform the data into the form you want. I'll probably start building one for cfb starting in June or July.

boboguitar said:

---

czar_iv said:

---

What are your inputs to your model?

---

RPI, SOS, Regular season win %, last 10 games win %, seed, and conference. Would have loved to add a bunch more but I did this in about 3 days in my free time. I'll start much earlier next year.

The data used is from 2000 - 2018 to train/test it.

---

I wonder if game site location plays any factor in outcomes.

Interesting. How much of a factor is experience of a team with upperclassmen and/or teams with tournament experience?

WestAustinAg said:

---

boboguitar said:

---

czar_iv said:

---

What are your inputs to your model?

---

RPI, SOS, Regular season win %, last 10 games win %, seed, and conference. Would have loved to add a bunch more but I did this in about 3 days in my free time. I'll start much earlier next year.

The data used is from 2000 - 2018 to train/test it.

---

I wonder if game site location plays any factor in outcomes.

---

I read something on 538 saying there is data that shows teams play worse the further they have to travel. I don't know how significant of an effect it is though.