2
$\begingroup$

currently I just met a case and Give the context below:

I got a 1000 (rows) x 6 (columns) data set.

The variables are Date, Hour, average temperature, average humidity, the sum of water consumed in one hour, average ph value in the water.

How should I build a model to predict the daily water consumption in the next few days?

The given test data set only has the average temperature, average humidity and the average ph value in the water. So I suppose I should mainly focus on these three variables to build models right?

At the same time, since this data set contains date and hour as variables, there are some missing value or some time line there is not data be recorded, shall I try to replace them back with KNN or other methods?

Best.

$\endgroup$

closed as too broad by mkt, kjetil b halvorsen, Michael Chernick, Peter Flom Mar 16 at 12:27

Please edit the question to limit it to a specific problem with enough detail to identify an adequate answer. Avoid asking multiple distinct questions at once. See the How to Ask page for help clarifying this question. If this question can be reworded to fit the rules in the help center, please edit the question.

0
$\begingroup$

you can try multiple models and cross validate the results. A simple approach would be to use ARIMA with exogenous variables. As you need daily prediction so if data is hourly, aggregate them.
For Arima , daily consumption would be univariate time series with other 3 variables as exogenious variables.
Before this do EDA on data and see if there is trend, seasonality etc. Check if external variables have any impact on consumption as consumption would be independent of temp, ph etc. Be careful using KNN for imputing missing values as it might give 0s where it is not able to impute values.

$\endgroup$
0
$\begingroup$

Decomposing Historical Data - Arimax versus linear regression discusses ARMAX models which is the preferred approach. It does require that the data is compete ... thus missing values need to be estimated/interpolated .

Some reasons why ......

1) single variable to be predicted 2) multiple suggested causals with possible lag effects 3) possible hourly,daily.monthly .. effects 4) possible holiday effects (?) 5) possible level shifts and trends 6) possible temporal response 7) possible parameter changes over time 8) possible error variance changes over time.

$\endgroup$

Not the answer you're looking for? Browse other questions tagged or ask your own question.