How to change a negative trend

[sohanasarah]

So I am generating a forecast for a dataset that always shows a negative trend. I removed outliers from the data and tried tuning the parameters. I also tried to clip the forecast result so it doesn't produce negative forecasts. I am not being able to change the downward trend. Do you have any suggestions on how to handle this situation?

Parameters I'm using:

params = dict(weekly_seasonality = False,
daily_seasonality = False,
yearly_seasonality= True,
seasonality_mode = 'multiplicative',
epochs=500,
learning_rate=0.01,
trend_reg=0.7)

[sohanasarah]

I have also added custom changepoints, set trend_reg_threshold to true and newer_samples_weight to 4,

[judussoari]

Hi @sohanasarah could you please share about the data you are looking at? I assume the green curve is the input time series, which would mean that you are trying to forecast 5 years into the future, which seems a lot to me. Does the same occur for a smaller forecast scope?
Also, you have not activated autoregression yet (using n_lags as a param). This accounts for the autocorrelation of the time series. I can imagine this to be what you are looking for