LifeStyle Targeting operator allows direct financial results forecasting compared to control group/average from raw transactions and questionnaire/impact data.

It generates and T-tests all offers’, customers’ and observations' hypothesis for all text, numeric and time field’s combinations, for example:

• [Sum(Cost) with 'Champagne' in 'GoodsName' on Morning] > 342.0
• [Word 'Director' in 'Title'] – Exists.
• [Offer: Words ‘40%’ in ‘GoodsName’] – Exists
• -115.82 + 109.40 * [Sum(Amount) with Word '12.5%' in 'Goods_name' On Friday]
  
• [Frequency Change (Number of visits half 2 - half 1)] – Not exists or < -800
• [Recency (days from last purchase)] – Not exists or < -35.5
• [ObservationDate] > 1305031258
• [Past Amount of Offer] >110
• [BirthYear] > 1987.0

Resulting decision tree (possibly with linear nodes in least squares mode) is used for forecasting in prognosis mode and can be saved as HTML for Excel use.

RCOMM 2012 Conference proceedings and sample RapidMiner process with 1% of NetFlix prize data are available (more in operator description).

Questionnaire input can be used for "flat" table data like traditional learning algorithms. Transactions’ input combines RFM, market basket and time series analysis.

Best characteristics are output as attributes, acting as feature generation and selection.

The extension is based on streaming LifeStyle Segmentation algorithm, revealing most financially profitable forecasting models with given statistical confidence supporting up to 4 billion unbalanced sparse attributes as decision tree with threshold and linear nodes. Analysis of millions of observations with millions of unique characteristics on a regular PC takes minutes to hours due to efficient disk and memory usage (patented WO2013055257).

Free version runs on Windows and has limitation of 100000 observations. Feel free to contact Skype:mDrobyshev or info@LifeStyleMarketing.org on any errors/suggestions, including other applications of the algorithm.