Missing value of time series variable

Most courses and online material don’t focus on how to impute missing values in different scenarios. Take a mean / mode of all other values is a very bad imputation technique in many scenarios. The underlying idea is that if a very small percentage of data has noise, a well tuned and not overfitting machine learning model will be robust to ignore that. However, if the percentage of missing data is high, that will not be the case. Furthermore, sheer dropping of missing values can not only lead to data loss, but also a different sample characteristics.
In case of time series and continuous data, spline interpolation can also be a technique to impute the missing data.