Works TS
PredictiveWorks. time series support has a strong focus on forecasting and prediction. Data are often sparse in time, non-stationary, carry seasonality pattern and trends.
A frequent requirement for time series techniques is that the data be stationary. This argument holds for the time series models supported here as well.
PredictiveWorks. responds to this fact and offers, in addition to model building, forecasting and prediction, many standardized plugins for preprocessing, seasonality & trend decomposition, and time series engineering.
Time Series Tasks
Preprocessing
PredictiveWorks. supports a variety of data preparation techniques to transform sparse time series data to a series of values uniformly sampled in time.
This includes aggregation, resampling, interpolation to fill missing values and more.
Decomposition
Time series data often carry seasonality pattern and trends and are non-stationary. Differencing is a frequent technique to make time series data stationary and ready for model building.
Seasonality and trends, however, have its own value. PredictiveWorks. supports decomposition of time series data into their seasonal, trend and remaining part.
This preserves and extracts meaningful information and also prepares data for model building.
Engineering
Many general purpose ML models such as clustering, regression and others, start from feature vectors and labels. To benefit from these ML models, time series data have to be transformed into (labeled) feature vectors.
PredictiveWorks. time series engineering supports time series embedding into higher-dimensional feature spaces.
Forecasting
PredictiveWorks. offers approved time series models as standardized pipeline plugins to look several steps ahead in time and forecast the values of a dependent variable.
Plugins for model building and forecasting share the same technology and make sure that trained and retrained time series models are immediately available for production pipelines.
The following time series models are supported: ARIMA, ARMA, AutoRegression, and MovingAverage.
Prediction
For many use cases, including demand prediction, forecasting what lies ahead in the future can be satisfactorily and easily solved after transforming it into a classification or regression problem.
PredictiveWorks. supports time series engineering to seamlessly integrate time series analytics with general purpose classification and regression.
In addition to this, PredictiveWorks. offers support for Random Forest Regression to predict the most probable value for the next point or interval in time.
Integration
PredictiveWorks. integrates Spark Time, a time series analysis library that has been open-sourced recently by Dr. Krusche & Partner. Its functional scope was externalized as standardized plugins for Google CDAP pipelines and can be easily combined with any other plugin.
