I am using the seasonal decomposition method from the statsmodels package to decompose time-series data into the trend, seasonality, and residual components. The issue is, I am trying to find a sound method for determining the best period (period of the series) value. I've seen some people choosing a period based on the p-value of the Augmented Dickey-Fuller test. They run it on the residual component to verify its stationarity, but this is not sufficient enough in my situation and I was wondering if there would be any other way for verifying period parameter selection? Note, that I've also used other stationarity/unit-root tests on residual data, but they fail at my task.