The Cornerstone of GiniMachine: AI for Predictive Analytics
At GiniMachine we are committed to developing technologies that create impactful and smooth experiences and serve as a key source of insights. We did not reinvent the wheel. We built GiniMachine on top of the strongest AI algorithms, adding some ‘secret sauce’ to make it more powerful.
GiniMachine is an ML-as-a-service platform that utilizes machine learning (ML) and makes AI meet the current business needs of our customers.
The platform enables financial institutions to seamlessly build, validate and deploy risk models at speed and scale. It is designed to automatically cover the end-to-end ML workflow: from data management to real-time predictions.
Why CROs need GiniMachine
As technology advances, risk management and data-driven decisions become both more accurate and more complex. Preparing data, checking for correctness, identifying outliers and empty records, and finally, training, building and validating models take too much time, money and manpower.
GiniMachine is designed to address these gaps through the end-to-end workflow.
For risk officers: GiniMachine simplifies the data science behind its system, making it easy to build high-quality models without a data scientist.
For data analysts: GiniMachine saves hundreds of hours of manual work and makes Big Data easy to understand.
The platform is a fully automatic and autonomous platform, requires no specialized knowledge in mathematical statistics or machine learning.
The software is a web-based application built with the combination of Java (web part) and Python (machine learning core). The solution uses a custom implementation of the decision tree ensemble method strengthened with a “secret sauce”.
GiniMachine is based on “supervised learning” algorithms, thus it uses historical data to build risk models specific for each customer.
The platform can work with any kind of data available to a business owner, both structured and “raw”. For example:
- Socio-demographic data
- Information from credit bureaus
- Transactional data
- Data from social media networks and
- Other data.
GiniMachine: Use Cases
You can use the platform for a very diverse set of applications. The platform can manage dozens of models across the company for a number of prediction use cases: Credit/Application scoring, Churn rate analysis, Marketing: X-Sell & Upsell, Collection scoring, In-depth data analysis.
The platform can analyze large amounts of data both financial and non-financial – with more granularity and deeper analysis – to isolate patterns important for a specific use case.
GiniMachine: Machine Learning Workflow
The system provides a scalable, and automated workflow to address a five-step process.
- Data management
- Model training and building
- Model validation
- Making predictions
GiniMachine works with historical data, without preliminary analysis and data pre-processing. You just need to prepare the .xls/.csv file format that contains the columns (attributes, parameters) and rows (records). Each record must have a status (outcome)*: 1- repaid, 0 – overdue.
Please note! Personal information (a full name, personal or file numbers, etc.) is not required. If any such data exists, the system automatically identifies and ignores it.
Model training and building
GiniMachine automatically divides input data into training and test records at a ratio of 70% to 30%. Based on the training records, the system builds a model and then checks it against the test records. The quality characteristics of the model and performance metrics (e.g. the Gini Index, ROC curve, etc.) are computed and combined into a model evaluation report.
Validate the model
You get a detailed report with all valuable insights and diagrams, every time the system builds a model.
- the calculated Gini Index for the test records;
- recommendations for the best cut-off value;
- a list of the most important attributes for the model;
- diagrams that reflect the results of the model performance.
Deployment and Real-Time Predictions
GiniMachine has end-to-end support for managing model deployment via a simple rest-API. You can instantly host the model or set it up on-premises, without a complex deployment process. Above all, every time the model is built and validated, it’s ready for risk calculations & real-time predictions.
- Fast, fully autonomous, and automated model building process. With a prepared dataset, it takes only 2-10 minutes to build and validate a risk management model. It saves hundreds of hours of manual work for risk officers and data analysts.
- High performance and predictive power. Typically, up to +15 Gini Index points compared to traditional models based on logistic regression.
- Ease of use – no special training is required to build a model.
- Built-in model evaluation and validation tools.
- Ability to use raw and big data. Ability to handle imperfect and missing data.