machine learning - Time Series forecast on Spark -

so i´m trying power consumption forecast time series data apache spark. sample of data :

03.01.15;22:30;236,25 03.01.15;22:15;240 04.01.15;16:00;243,775 

and on 2 year. have observations every 15 minutes

what best way predict power consumption ?

i try linearregression, decision trees etc. huge mses (788). try pass months, days, hours, minutes onehotencoder. try forecast weeks etc.

means of data year, month, day:

[2014.3996710526321,5.726973684210525,15.713815789473673] 

variance of data year, month, day:

[0.2403293809070049,10.218579294199253,77.46326844706495] 

test mean squared error

788.2397552290726 

if pass values direct labeledpoint(236.25, 2015.0,1.0,3.0,22.0,30.0)) mse goes 1280.8. if pass model 1 observation per day being max value not to.

but if use knime , try example time series data not take dates , time in consideration, instead lagged power consumptions each observation.

i see cloudera has library time series not understand why need it.

can describe process of doing forecast on time series data ? @ end want input date , time , prediction.

i have multiple questions issue, let me try work you've given me.

first, let's generalise problem.

you have data in form <timestamp>, <value>. given data collected every 15 minutes 2 years, have sample size of (4 x 24 x 365 x 2) 70080 observations (rows)

and let's want develop regression model predict behaviour.

first things first, need partition dataset training , test sets. develop model using training set , test model on test set. suggest 9:1 split.

q1. kind of approach have taken?

mse calculated using predicted values against actual values in test set.

q2. mention mse - have no idea how calculated (only 3 data points provided) or range of values working with. can please affirm how error calculated , minimization criteria is?

if linear regression fails (cannot predict movements succesfully), may case model simple - try using mlp or combine regression model bayesian model (as power consumption continuous function).