Daily timeseries decomposition into seasonal, trend, remainder?

I have time series for each day, that captures the time(s) of the day where a certain event $$E$$ happens (or alternatively, when it certainly isn't happening). This looks like the following:

time | state | monday | tues | wed | thurs | fri | sat | sun
28810 | True | 1     |  0    | 0  |  0    | 0   |  0   | 0
28890 | False | 1    |  0    | 0  |  0    | 0   |  0   | 0
28925 | True | 0     |  1    | 0   |  0    | 0    |  0    | 0
...


where time signifies the time of day as seconds elapsed since midnight. I have then transformed the time component such that, $$s_t = 24*60*60 = 86400$$, $$sintime = sin(\frac{2*\pi*time}{86400})$$ $$costime = cos(\frac{2*\pi*time}{86400})$$ and dropped the time column.

I need to feed the data into a classifier but this data does not work well with a neural network. How can I decompose each daily time series into several components to create more features (whether with or without the sin and cose pair transform)? Note that I want daily trends because events happening throughout each day affect each other, so it might be beneficial even though there are only around 200-300 data points for each day over the course of ~3 months.

I have found this and this but I'm none the wiser.