MoneyCare implemented instruments of machine learning

On June 19, 2017 the company Columbus announced creation of a forecasting model on the basis of a cloud service Microsoft Azure Machine Learning by request of the loan broker MoneyCare. The technology helps to estimate the probability of affirmative answer of bank on a credit request.

Project Tasks

MoneyCare decided to reduce the number of biographical particulars to credit requests, minimum necessary for the purpose of increase in conversion. The decision on creation of a forecasting model of probability of affirmative answer of bank is at the same time made. The company charged a task - to define the minimum data set and to create a prototype, to consulting company Columbus.

Use of cloud solutions allows to unroll quickly desirable infrastructure with the minimum investments. Cloud computing opens the wide field for experiments and allows to select the most effective options of the innovative solutions. For example, to use machine learning for forecasting, without investing in development of computing powers or analytical tools. Evgeny Lebedev, the head on business development of cloud solutions of Columbus company

Selecting the platform of machine learning specialists of MoneyCare stopped on a cloud service of Azure Machine Learning.

Exact forecasting – a key step to success in financial market. Microsoft Azure Machine Learning provides an interactive visual working space, simplifying creation, testing and the most important deployment for the subsequent use of models of predictive analytics. Tatyana Delyagina, promotions manager of Data Insight of Microsoft company

Project Progress

At the first stage the qualifier prototype in Azure Machine Learning which task - selection more than 60% of requests for the credit with approval probability more than 80% was created. The used methods of machine learning:

• discriminant analysis,
• regression analysis,
• clustering,
• classification on the basis of divisibility (SVM, ANN),
• algorithms of reduction of dimension (PCA).

The second part of the project - training of employees of the customer in the principles of work a joint development environment for improvement of a prototype. This stage included consultation on setup of models into Azure Machine Learning, to standard problems of machine learning, determination of further actions.

Project Results

Despite popularity of a subject, there is not a lot of implemented projects on machine learning. First, the bad quality of initial data affects – there is no information which can be used for forecasting, often just. The second problem – staff deficit. It is possible to design any prototype, a question in the one who will use it then. Creation of a forecasting model by machine learning was our task not just - it was necessary to train in the customer of the specialist of Data Science capable to develop model, to test on it new hypotheses and to adapt parameters to the changing environmental conditions. I think, at us it turned out. Roman Mikhaylov, director of practice of information and analytical systems of Columbus company