Why ModelCraft™?


ModelCraft™ is an automated machine learning and component-based modelling tool that allows simplifying and industrializing the development and validation of advanced analytics models

ModelCraft™ has intuitive components for data processing, advanced analytics model development and interpretability. It allows users to connect to any data repository and to generate modelling workflows in a visual manner and without the need for programming. 

ModelCraft™ thus responds to industry needs such as: 

  • Incorporating advanced machine learning algorithms into the modelling process in a simple and intuitive way, such as gradient boosting, elastic nets, ensemble models, random forests, support vector machines or neural networks, among others.
  • Industrialize and accelerate the modelling process, reuse codes and reduce the time spent on routine and repetitive tasks.
  • Facilitate access to advanced modelling techniques for non-specialist teams.
  • Improve the performance, traceability and comparability of models.
  • Enable the interpretability of models and automate the generation of their documentation.
  • In short, to improve the model development and validation processes, and to enable the generation of advanced challenger models.
  • In addition, ModelCraft™ has been developed in the cloud and natively uses the capabilities of auto-scalability and parallel computing.

    Thanks to this, ModelCraft™ is a tool to accelerate the transition to a component-based modelling, which is more efficient, more interpretable, more accessible and with lower model risk.

     

     

     

    Functionalities of ModelCraft™


    ModelCraft™ has a library of predefined components that allows developing the complete modelling lifecycle 

     

    Input, connection and download 

  • Input data reader with different formats and delimiters
  • Model reader in .pkl format
  • Database connection components (cloud, RDS)
  • Download of models, files and reports
  •  

    Sampling 

  • Sampling algorithms 
  • Obtaining training, test and validation samples 
  • Algorithms for processing highly unbalanced data 
  •  

    Visualizations 

  • Model comparison and selection of the optimal model 
  • Exploratory data analysis 
  • Representation of histograms, pie charts, etc.
  • Analysis of database structure 
  • Calculation of descriptive statistics 
  •  

    Performance and prediction 

  • Mapping of criteria to training samples, validation and testing 
  • Calculation of metrics and KPIs 
  • Out-of-time analysis and backtesting 
  • Predictions 
  • Model validation 
  •  

    Database management 

  • Simultaneous connection to different data repositories in various languages (postgresql, mariadb, mysql, oracle, terminal, etc.) 
  • SQL data processing (generation of tables, data manipulation and fetching, table traversal, etc.) 
  • Import and export of files 
  •  

    Feature engineering

  • Binning algorithms 
  • Correlation analysis 
  • Variable selection and elimination 
  • Missing variables analysis and imputation 
  • Outlier analysis and treatment 
  • Cálculo de weight of evidence.
  • Encoding techniques 
  •  

    Model training 

  • Hyperparameter search techniques 
  • Ensemble methodologies 
  • Genetic algorithms 
  • Training of supervised and unsupervised models
  • Neural networks 
  •  

    Interpretability 

  • Variable importance analysis
  • Branch & bound algorithms
  • Surrogate model building
  • Univariate and bivariate analyses
  • Interpretable trace of the entire development process
  • Automatic generation of a complete model report
  •  

    Automated machine learning

  • Generation of user-created components, including free code, in real time 
  • Determination and optimization of the critical path between components 
  • Encapsulation of components in three levels of depth 
  • Access to Python and R code generated by each component 
  •  

     

    Features of ModelCraft™

    ModelCraft™ enables the application of machine learning techniques in an intuitive and robust way, and it is deployed in a simple, flexible and scalable technology

    User


    Ease of use:

    The ModelCraft™ interface allows the visual buildup of modelling flows and gives access to all parameterization in an organized way. 

    Personalization

    Users can create their own components, design their own modelling flows and access the generated Python and R codes. .

    Explainability

    ModelCraft™ can generate an automatic summary of the model, its variables and its complete development process at the click of a button, describing the decisions made by the user (e.g. the model's hyperparameters) and by the algorithms (e.g. the exclusion rationale for each variable). 

    Evolution

    ModelCraft™ is constantly evolving and adapting to the needs of the industry; for example, the probability of default (PD) calculation has already been added to the ModelCraft™ component library, and LGD and CCF will be added shortly. 

    Traceability

    The interface offers full traceability of the developments made, allows access to all the code, and provides logs and traceability functionalities. 

    Optimized code

    Each component in ModelCraft™ has been optimized to maximize computational power, using libraries available from the most advanced programming environments. 

    Integration

    ModelCraft™ integrates natively with two proprietary Management Solutions: ModelCraft Integrator™, for putting models into production through a model and rule manager that interacts with operational systems, and ModelCraft Monitor™, for periodic monitoring and backtesting of models. 

    IT


    Deployment flexibility

    ModelCraft™ can be deployed as an on-premise tool or as software-as-a-service (SaaS) in the cloud, according to customer needs.

    Scalability and parallel computing

    ModelCraft™  is scalable and takes full advantage of parallel computing capabilities, allowing the processing of large volumes of data and executions with a high computational load. 

    Easy deployment

    The designed architecture allows ModelCraft™ to be deployed in a very short time, including an analysis of customer needs, their configuration preferences and necessary customizations. .

    Reduced support

    ModelCraft™  requires reduced IT support, and maintenance includes the deployment of new releases. 

     

     

     

    Structure of ModelCraft™


    ModelCraft™ has two work areas: the data management panel and the modelling panel

    The data management panel allows users to connect to any database, and extract, merge and load tables with SQL language. The modelling panel allows, through components, to perform exploratory data analysis, prepare data for modelling, and train and interpret advanced machine learning algorithms.

    Data management

    • Database management
    • Simultaneous connection to different data repositories in different languages (postgresql, mariadb, mysql, oracle, terminal, etc.)
    • SQL programming

    Modelling

    • Data processing and EDA
    • Modelling components and advanced machine learning algorithms
    • Interpretability and automatic reporting components

    In turn, the modelling components are structured in three levels of depth: from level 0 (minimum executable task, such as outliers detection or model training) to level 2 (aggregation of components to build an end-to-end modelling process)

    Modelcraft Structure


    IT architecture 


    ModelCraft™ has been developed cloud-natively, using a Kubernetes-based architecture that enables rapid deployment of the component library and a visual programming frontend.

     

    Arquitectura

    The ModelCraft™ architecture is focused on different stacked component libraries, allowing the use of the components as microservices, and includes a user, storage and execution manager.

    Credentials 


    Management Solutions has extensive capabilities and references in the field of advanced modelling, combined with in-depth knowledge of technology and business processes

    Management Solutions provides differential value in the field of advanced modeling: 

    • It has extensive experience in the development, validation and implementation of models and algorithms applied to multiple business areas in different industries and sectors.
    • It has quantitative capabilities and resources with a high degree of knowledge and specialization (two thirds of its staff are STEM).
    • Specifically, it has teams of data scientists who carry out projects in artificial intelligence and machine learning, as well as an R&D team dedicated to the practical application of AI applied to management.
    • It has developed a Chair in Big Data and Analytics together with the Universidad Politécnica de Madrid, where it develops lines of research on modelling techniques and artificial intelligence. This has allowed it to develop ModelCraft™ and keep it up to date with the latest advances in modelling and computing.

    Thanks to these differential values, Management Solutions implements ModelCraft™ in all its industries and in all the countries where it operates.

    ...

    Financial entities







    ...

    Energy & Utilities







    ...

    Telecommunications







    ...

    Other sectors: manufacturing, pharma, logistics and transportation, tourism, professional services, education, construction and infrastructure, government

    Management Solutions

    Management Solutions is an international consulting firm, focused on business, finance, risk, organization, technology and process consulting.

    Management Solutions currently has a multidisciplinary team (functional, mathematical, technical and systems integration) of more than 2,700 professionals who carry out their activities through 33 offices (16 in Europe, 16 in the Americas and 1 in Asia), from where they provide recurring services to clients operating in more than 40 countries in four major geographic areas: Europe, Asia, Africa and Latin America.

    We create value propositions, committing ourselves to their effective implementation, exceeding the expectations of our clients and becoming a trusted partner

    Management Solutions' differentiating factor lies in its in-depth knowledge of the businesses in which its clients operate, and its high level of sector and functional specialization.

    In order to meet its clients' needs, Management Solutions has structured its practices by industry and business line, grouping a wide range of competencies.

    For more information: www.managementsolutions.com

    For further information, please contact:

    Manuel Ángel Guzmán
    Partner at Management Solutions
    manuel.angel.guzman@msspain.com
    Javier Calvo
    Partner at Management Solutions
    javier.calvo.martin@msgermany.com.de
    José Manuel Suárez
    Partner at Management Solutions
    jose.manuel.suarez.fernandez@msspain.com
    Segismundo Jiménez
    Manager at Management Solutions
    segismundo.jimenez@msspain.com