Predict next date of incident using R from dates only I have name of incident and date, which is not univariate. I need to do prediction/forecasting for upcoming event based on date. I have tried to do it by using average of date difference between consecutive dates. Also tried to do it by using rollmean for latest trend.
> head(df1$Date)
[1] "2014-01-20" "2014-01-22" "2014-03-10" "2014-04-10" "2014-04-15" "2014-04-15"

I was wondering is there any other way to do it?
 A: Check out a technique called 'landmarking' in survival analysis. Basically looking at the historical data from an ex-ante perspective. If you are trying to forecast two quarters ahead from now. Then you looking at the historical data in a way that the covariates information is is two quarters ahead of the results. 
A: Working with day differences between dates.
If you know the data to be stationary and not auto-correlated (i.e. it doesn't trend, and the occurrences happen roughly at a constant rate), then consider modelling it as an arrival process. I.e. fit a Poisson distribution to it. Then E(X) gives you your forecast, and you can also show confidence intervals and all that jazz.
If you know the data to be auto-correlated and/or not to be stationary and you're not particularly experienced in time-series forecasting, then I'd suggest starting out with exponential smoothing. R has very easy to use packages for this. If you want confidence intervals, you can fit a distribution to the residuals or bootstrap them.
