TIM Introduction

Increase your creativity!

Most statisticians or dataminers are using Statistical Tools that are limiting their creativity and productivity. As a result of these limitations, statisticians are usually spending a huge amount of time to:

•   reduce the number of rows of their datasets (sampling)
  
  
•   reduce the number of columns of their datasets
  
  
  • It's very common to use some business sense to eliminate "a-priori" a large number of your variables. This is not a good idea. You should let TIM decide if a variable must be removed or not. An another "bad habit" is to delete variables with many missings. Very often the target size is a few percent, in this situation, variables with 90% missings are perfectly legitimate.
    
    

    Real Business Case: In the B2B field, a few years ago, a customer from a famous bank came to see me because his statistician team had problems to obtain a good lift on their dataset. They had only a lift of 2.5. This dataset contained 3000 columns (the columns contained information extracted from company balance-sheet for several years). Height minutes after receiving a copy of this dataset, I obtained a new model with a lift of 4. The model was based on 15 variables. Most of the variables used by the new model were not even considered for analysis by the statistician team: they were eliminated "a priori" using an "adhoc" heuristic. The eliminitation was based on a test exploiting the curve of the density of the target VS the value of the variable. If this curve is oscillating too much (not strictly increasing or decreasing), the variable is directly removed. This eliminitation was performed because the statistical software they were using was not able to handle practically more than 300 columns.

• 
  
•   clean their datasets so that no outliers remains
  
  
•   recode or transform the variables to:
• - obtain normal distributions.
• - improve the lift of their model
    For example: it's very common to create new variables that are the LOG of monetary values


•   prevent over-fitting
  
  
•   try to find the most appropriate modelization procedure (trees, neural nets, logistice regression...)
      spending countless hours tuning the hundreds of parameters available.
  
  
•   wait for the statistical tool to finish its computation
  • Furthermore, classical datamining tools can simply NOT analyse datasets of 20 millions rows and 20 thousands columns (... or they could do the analysis but the computation lasts for 2 weeks ).

All these concerns no longer exist when working with TIM. You can focus on the business sense of your data to directly create the best models.

Explore new area of knowledge!

If you are an experienced statistician that likes to create complex, high perfomance models (predictive models or segmentation models), you will be more than happy to discover the new possibilites that TIM offers you. In opposition to other concurrent datamining softwares, TIM is not a "black box": You can change the parameters of all the algorithms inside TIM to obtain the highest performances. Default values are fine in most cases but advanced users are able to produce even better models using their expertise in the statistical field. Furthermore, there is no "impossible to understand" parameters inside TIM: you don't have to read 1000 pages of statistical text to understand one parameter or one strange KPI. If you understand what a lift is, you are "good to go"! With TIM you can concentrate on understanding your data and not concentrate on understanding the software that analyse them . Furthermore TIM opens completly new and exciting ways of extracting knowledge out of your data. The quality of your work will simply be boosted beyond proportion!


                                   

TIM for binary classification

The Binary Ranking System of TIM offers you the following capabilities:

•   Produce Ranking & Scoring for many kinds of different business cases.
  
  
•   Produce purchase probability for many kinds of different business cases
  
  
•   Produce probability of defaults for many kinds of different business cases (to avoid bad payers, bankrupted clients, cost consumers, fraud detection,...)
  
  
•   Produce probability of churn for many kinds of different business cases.
  
  
•   Produce a data quality analysis
  
  
•   Produce advanced customer profiles including dozens of nice colorfull graphics. A better understanding of your clients allows you to fulfill their expectation and to optimize the adequacy between you product line and their needs.
  
  
•   You get a thoughtful real-time analysis of the ROI of your campaign (this is unique to TIM)
  
  
•   You get a segmentation analysis module to easily explain your customer segments
  
  
•   You get a "Commercial Potential Analysis" of your data so that you can allocate optimally your commercial resources (this is unique to TIM).
  
  
•   Easy industrialization of even the most complex tasks.
  

TIM for continuous prediction

The Continuous Prediction System of TIM offers you the following capabilities:

•   Estimate the exact amount of money a lead will give you.
  
  
•   Estimate the "share of wallet" of your customer/lead (this is unique to TIM).

TIM for Multi-class prediction

The Multi-class Prediction System of TIM offers you the following capabilities

•   Estimate in which class a customer belongs
  
  
•   As a bonus, you can easily do boosting, bagging and Feature Selection as the Multi-class voting system. Working with sets of 5000 voting models is easy.

TIM for segmentation analysis

The Segmentation System of TIM offers you the following capabilities:

•   Detect (multivariate) Outliers and non-valid data
  
  
•   segment you customer base using the most advanced segmentation methodology
  
  
•   describe each segment from a business-point-of-view
  
  
•   a powerfull visualization engine based on the latest advances in 3D hardware accelerated rendering.

TIM Unique habilities

•   datasets of any size
•   guarantee to obtain the best model on any given real-world dataset
•   complete profile analysis of your target customers.
•   Estimation of the stability of your model: you get confidence intervals direclty on the lift chart.
•   share of wallet estimation
•   unmatched speed,
•   short time to market
•   realtime campaign optimizer,
•   customer profile directly actionable from a business perspective,
•   ...