Amra.

Data Scientist/Analyst

Hello, I'm
Amartuvshin

Experienced Data Analyst/Scientist skilled in Python, machine learning, and backend development. Proven track record in leveraging data insights to drive business decisions and optimize processes.

Download CV

5

Years of
experience

8

Technologies
mastered

1

HackAIthon
conquered

#1

Project Team
at Khanbank 2023

Involved projects

Software Development Projects

The Risk Management System

An internal web application used by Khanbank employees to monitor and evaluate customer-related risk factors, including behavioral scores. Additionally, the system provides functionalities for estimating corporate customers loan limits. It also serves as a primary tool for recording and managing actions taken during the debt collection process.

Digital Loan Project

This project involved transforming traditional retail loan applications into a fully online process. Historically, calculating customer loan limits relied on complex formulas in Excel across various branches. My role was to ensure that the new online loan calculation system produced results consistent with traditional Excel calculations. To achieve this, I developed an automated testing process using over 1,000 test cases. This was implemented using Python libraries such as OpenPyXL, Pandas, and NumPy, along with tools like python requests to streamline and validate the accuracy of loan calculations efficiently.

Machine Learning Projects

Business Customer Identification Model

This classification model leverages transaction data (including amount, frequency, and transaction description) to identify customers engaged in business activities and determine their business sectors. Using XGBoost, the model was developed to classify customers effectively based on their transaction patterns. It plays a critical role in the banks online loan approval process for business loans, enabling more targeted and efficient lending decisions. I served as the lead data scientist on this project, overseeing the development and deployment of the model to ensure its accuracy and reliability.

Family Linkage Model

This model identifies bank customers relatives and spouses by analyzing transaction data patterns. Using a classification approach with XGBoost, the model successfully discerns familial relationships based on shared transactions, frequency, and other transaction characteristics. The insights derived from this model help in providing personalized services and understanding customer networks as well as debt collection. I played a pivotal role as the lead data scientist in this project, overseeing the development, training, and implementation of the model to ensure its accuracy and effectiveness.

Deposit Balance Prediction

This project aimed to predict future deposit balances for SME and corporate customers using various time series forecasting techniques. Recognizing that customer deposits are crucial to the banks operations, we developed a predictive tool to identify periods when customers are likely to have low balances, allowing the bank to proactively engage with them. We evaluated multiple time series forecasting methods, including ARIMA, XGBoost, and LSTM. After thorough comparison, LSTM was selected due to its superior performance in capturing complex patterns in the data. As the lead data scientist on this project, I was responsible for model development, evaluation, and implementation.

Behavioral Scoring Model

This model is designed to predict the likelihood of retail customers repaying their loans based on their transaction data. The project was developed in collaboration with Visa Corporation, leveraging their expertise in behavioral analytics. My role involved coordinating between the Visa team and Khan Banks internal team during the early stages of the project, ensuring effective communication and alignment of goals. This model enables the bank to better assess customer credit risk and make informed lending decisions.