YModel

One-click Automated Modeling

Manual modeling

  • Data exploration?
  • Noisy data?
  • Time characteristics?
  • Complicated business requirements?
  • How to evaluate a model properly?
  • High cardinality variable?
  • Non normal distribution?
  • Standardization?
  • LR, RF,GBDT…..How to choose the right algorithm?
  • Prolonged project cycle?
  • Missing values?
  • Parameter configurations?
VS

YModel Auto-Modeling

Fast, effective, fully-automated, one-click modeling!
Build great models in the absence of data scientists

Traditional modeling process

Data scientists + Heavy manual labor
  • Data input
  • Modeling data preprocessing
  • Manual modeling
  • Model performance
  • Model output
    • Variable type identification
    • Data processing load control
    • Data quality report auto-output
    • Exception handling
    • Missing value handling
    • High cardinality variable handling
    • Data smoothing
    • Numeric variable auto-filtering
    • Derived variable addition
    • Get important variables
    • Optimize model parameters
    • Select modeling method
    • Auto-select important variables
    • Auto-optimize model parameters
    • Auto-optimize modeling method
    • AUC
    • GINI
    • MSE
    • LIFT
    • KS
    • RECALL RATE...z
  • Data input
  • Auto data preprocessing
  • Intelligent modeling
  • Model performance
  • Model output
    • With YModel Auto-Modeling, data scientists & manual labor out, quality and stable models in

    YModel Auto-Modeling process

    1.Variable type identification

    2.Variable processing

    3.Auto-preprocessing & -modeling

    4.Model evaluation metrics

    YModel Auto-Modeling architecture

    Why YModel?

    Fruit of lifelong data mining experience

    Decades of practical experience in business modeling and data mining; Modeler and manager of data mining projects in the banking and insurance industry; SAS competition winners.

    Innovative & powerful modeling tool

    Penetrating perspective on mathematics; extraordinary software implementation; Industry-leading high-performance big data processing technology.

    Case1: Personal loan default forecast

    Target

    • Build loan default model to evaluate users default probability
    • Give reasonable credit line to users
    • Allow line employees and managers to select data for modeling based on experience, and press forward the use of models among them
    • Increase rate of default capture

    Pain points

    • Find a reasonable data dimension
    • The impact of high cardinality and nonlinear problems on modeling
    • Select the right model or model combination
    • Avoid overfitting due to less positive samples

    Modeling performance comparison

    Auto-Modeling Manual modeling
    Modeler 1 1
    Time 5 minutes（data preprocessing + modeling） 2 months
    Model 1 1
    Data volume 100000+ / 28MB 100000+/ 28MB
    Model AUC 0.9728（test set 0.965） 0.957

    Model performance（test set）

    Case 2: Precision marketing for financial products

    Customer group 1 Customer group2 Customer group3 Customer group4
    Modeler 1 1 1 1
    Model 13 13 13 13
    Modeling time 1.5 hours per model 1.5 hours per model 1 minute per model 2 minutes per model
    Data volume 1.34 million 1.55 million 6400 12000
    Cumulative lift Cumulative capture rate AUC
    First 5% 14.4 72% >0.9
    First 10% 9.4 94%
    First 15% 6.3 94.5%
    First 20% 4.8 96%

    The current purchase rate of the financial product is 1.72%

    1. The purchase rate of the first 5% data using the model is 14.4 times higher than that without the model. That is, for every 100 selected customers, 24.77 transactions can be completed. It is far higher than the average of 1.72 transactions per 100 customers.
    2. 72.0% of the target customers can be captured from the first 5% of the data grabbed by the model. 96.0% of the target customers can be captured from the first 20% of the data grabbed by the model.

    Auto-modeling VS Manual modeling

    Number of models duration Number of modelers
    Auto 50-60 2 weeks 1
    Manual Not suitable for mass modeling 1 week ~2 months per model
    (The actual modeling time is uncontrollable because it depends on the degree of complexity of modeling and the skills of modelers)    		 A number of

    Features of YModel Auto-Modeling

    More automation , less human labor
    Auto-Modeling

    Efficient

    Non data scientists

    Low cost

    Performance improvement

    High accuracy

    YModel Auto-Modeling changes data mining mode to user-orientation and instant modeling

    Data Mining Course