I have a dataframe containing various vital signs (dbp, sbp, mbp, hr (heart rate), bis, exp_sev, insp_sev (inspired/expired sevolfurane) collected over time for multiple patients. The acquisition times for these signals vary, resulting in null values at different intervals: bis is acquired every second, so it has very few null values. dbp, sbp, and mbp are acquired every 2 seconds, resulting in null values every other second. exp_sev and insp_sev are acquired every 7 seconds, resulting in null values every 7 seconds. Given this scenario, I am looking for the best approach to handle these null values, considering I want to run both linear regression and LSTM models on the data (in python). Questions: Should I impute the null values, and if so, what method would be most appropriate (e.g., interpolation, forward fill, backward fill)? Is it better to leave the null values as they are since the test set will also have them? Additional Information: The data is in a time series format, and each row corresponds to one second in time. I want to preserve the temporal integrity and relationships between the different vital signs. Any insights or recommendations would be greatly appreciated! Continue reading...
