Facebook prophet actual and prediction values varies widely

I'm presently using Facebook Prophet to forecast bitcoin. It's very nice but I noticed the wide difference between the prediction and the actual values are very wide, The are points where the prediction and actual values differ by more than 15%. I would like to fit this for better accuracy. Please help. I'm using 5 minutes data.

This is the code

library(prophet)  library(lubridate)  library(ggplot2)    mydata <- read.csv('~/Arbitrage/Mar-20-2021/5m_data/BNBUSDT/BNBUSDT.csv')    ds <- mydata$Date  y <- mydata$Close  df <- data.frame(ds, y)  qplot(ds, y, data = df, main = "Start working")  m <- prophet(df, changepoint.prior.scale=0.01)  future <- make_future_dataframe(m, periods = 300, freq = 60 * 60)  forecast <- predict(m, future)  plot(m, forecast)+add_changepoints_to_plot(m)  dyplot.prophet(m, forecast)  prophet_plot_components(m, forecast)  

https://stackoverflow.com/questions/66739478/facebook-prophet-actual-and-prediction-values-varies-widely March 22, 2021 at 10:07AM