• A
  • A
  • A
  • ABC
  • ABC
  • ABC
  • А
  • А
  • А
  • А
  • А
Regular version of the site

Modelling and Forecasting in Insurance Using Machine Learning and Neural Networks

Student: Magomedov Kismat

Supervisor: Yuliya Mironkina

Faculty: Faculty of Economic Sciences

Educational Programme: Economics and Statistics (Bachelor)

Year of Graduation: 2020

Artificial intelligence (AI) and big data penetate deeper and deeper into almost all areas of human activity. AI combines segments such as Computer Vision, Natural Language Preprocessing where neural networks are used mostly classic machine learning tasks that involve working with tabular data, time series prediction using neural networks and others. The latter task will be discovered in this study, in addition to the problem of predicting damage under an insurance policy. One of the areas where modern methods of data analysis (AI) are actively penetrating is insurance. The time series prediction (from the end of 2007 to 2012) will be performed using both classical methods (SARIMA, Holt-Winters model) and more modern ones like neural networks (Recurrent Neural Networks, sequence to sequence architecture (seq2seq)), gradient boosters, and random forest). The initial time series is non-stationary, and there are prerequisites for a structural shift. The next task will be to build the models for predicting damage based on the characteristics that describe the insurance policy. For example, there will be features like information about the driver (gender, age, experience) and about the vehicle (year of production, cost). In this problem, the classic tool that is used in the real insurance business is GLM-generalized linear models. In addition, Ridge and Lasso regression models, boostings, and random forest will be trained. The models will be configured on the validation set by grid searching for optimal parameters. The resulting models will be compared using such metrics as MAPE (mean absolute percentage error) in the time series prediction problem and MAE (mean absolute error) in the insurance policy damage prediction problem. The result of this work is to select the best models based on target metrics on the test data set and compare them with each other. The main message of the paper is that modern algorithms of machine learning and neural networks can replace the classic models that are currently used in the insurance business.

Student Theses at HSE must be completed in accordance with the University Rules and regulations specified by each educational programme.

Summaries of all theses must be published and made freely available on the HSE website.

The full text of a thesis can be published in open access on the HSE website only if the authoring student (copyright holder) agrees, or, if the thesis was written by a team of students, if all the co-authors (copyright holders) agree. After a thesis is published on the HSE website, it obtains the status of an online publication.

Student theses are objects of copyright and their use is subject to limitations in accordance with the Russian Federation’s law on intellectual property.

In the event that a thesis is quoted or otherwise used, reference to the author’s name and the source of quotation is required.

Search all student theses