Getting straight line while creating ARIMA model

Question

I have a Fan Speed (RPM) dataset of 192.405 Values (train+test values). I am training the ARIMA model and trying to predict the rest of the future values of our dataset and comparing the results.

While fitting the model in test data I am getting straight line for predictions

from sklearn.model_selection import train_test_split 
from statsmodels.tsa.arima_model import ARIMA

dfx = df[(df['Tarih']>'2020-07-23') & (df['Tarih']<'2020-10-23')]

X_train = dfx[:int(dfx.shape[0]*0.8)] #2 months
X_test = dfx[int(dfx.shape[0]*0.8):] # rest, 1 months

model = ARIMA(X_train.Value, order=(4,1,4))
model_fit = model.fit(disp=0)
print(model_fit.summary())

test = X_test
train = X_train

What could i do now ?

Answer 1

Your ARIMA model uses the last 4 observations to make a prediction. The first prediction will be based on the four last known data points. The second prediction will be based on the first prediction and the last three known data points. The third prediction will be based on the first and second prediction and the last two known data points and so on. Your fifth prediction will be based entirely on predicted values. The hundredth prediction will be based on predicted values based on predicted values based on predicted values … Each prediction will have a slight deviation from the actual values. These prediction errors accumulate over time. This often leads to ARIMA simply prediction a straight line when you try to predict such large horizons.

If your model uses the MA component, represented by the q parameter, then you can only predict q steps into the future. That means your model is only able to predict the next four data points, after that the prediction will converge into a straight line.