0

I am applying mixture modeling, and I have plotted the result of the AIC for different number of components. I know that the lower the better, but in this case I am doubting about what is really the best. What do you suggest and how can I best substantiate this? link

By myself I thought that 5 components would be the best. But I don't know What you guys think and especially how to substantiate.

daantheboss
  • 43
  • 5

1 Answers1

0

You have to calculate the gradient of the BIC scores curve. When the gradient of BIC scores gets almost constant it means that there is no more gain by adding more components.

K. Peltzer
  • 326
  • 3
  • 7
  • You mean really the BIC? Or is the gradient of only the AIC sufficient enough? – daantheboss Nov 06 '19 at 13:31
  • @daantheboss AIC and BIC are both penalized-likelihood criteria. Their only difference in practice is the size of the penalty. BIC penalizes model complexity more heavily. The only way they should disagree is when AIC chooses a larger model than BIC. It's good to use AIC and BIC together in model selection. AIC is better in situations when a false negative finding would be considered more misleading than a false positive, and BIC is better in situations where a false positive is as misleading as, or more misleading than, a false negative. – K. Peltzer Nov 06 '19 at 15:12