I have done behavioral experiments within 24 hours (9 time points). To show the rhythm, I applied cosinor analysis, but some of the data obviously overfitted. Still confused that if Cosinor model is suitable for 24-hour's data.
Tricky to know without knowing more about how it's overfitted, the distribution of the data and the cosinor method that you used. If you're working with counts, then a cosinor method which allows for a link function to do poisson regression may be more helpful (see RhythmCount or make your own with glm() in R).