Vusal Babashov

Vusal Babashov

Senior Scientist in Operations Research and Data Science

Sitemap

A list of all the posts and pages found on the site. For you robots out there is an XML version available for digesting as well.

Pages

Posts

Apache Spark

less than 1 minute read

Published:

Apache Spark has gained a lot of popularity recently for it’s vestatility and speed. It is among the Forbes’ top 10 best Data Analytics and BI platforms and tools of the 2020. I have started an online course on Apache Spark ecosystem and databricks which builds on top of Spark adding reliable and performance data pipelines.

Welcome!

less than 1 minute read

Published:

This is my humble website where I post opereations research and data science related stuff. Check back often as I keep updating it with cool projects, and ideas on advanced analytics.

portfolio

Do We Have To Wait So Long?

Published:

In the current clinical practice, priority-specific wait time targets are typically determined by the consensus of medical specialists and healthcare administrators. The problem with this rationale is that it does not consider the efficient use of clinical resources and the patient volume associated with each class. The aim of this method presented here is to determine wait time targets in a multi-priority patient setting in a systematic fashion that both respects clinically acceptable wait time targets and considers clinic size and demand distribution. This approach utilizes predictive, prescriptive and descriptive analytics. More specifically, simulation, deep neural network, regression, and inverse optimization approaches are used.

Bank Note Demand Forecasting

Published:

A central bank of Canada runs a distribution network and maintains an inventory of bank of notes at regional distribution points for multiple types of denominations. Both shortage and capacity overage of notes at the regional inventories need to be avoided. The goal of this research exploration is to come up with a forecasting model that can help the Bank Note Distribution System (BNDS) operations team to provide right amount of notes in the right place at the right time. Implemented models include classical time series approaches such as STL decomposition, TBATS, Dynamic Harmonic Regression (i.e., Arima with harmonic terms) and deep neural network approaches such as Multi-layer perceptron (MLP), Long-Short Term Memory(LSTM) and Light Gradient Boosting Method (LightGBM).

When Do You Need to Have Your Next Appointment?

Published:

In many healthcare systems patients require multiple visits to a healthcare provider. In general, the first visit is known as the consult visit and all the subsequent visits are known as the follow-up visits. The latter typically occur according to predefined booking guidelines. A Markov Decision Process model is used to efficiently allocate available capacity to consult and follow-up visits in a dynamic fashion. To solve this model, a Linear Programming approach to Approximate Dynamic Programming (ADP) is used. The characteristics of the approximate optimal booking (AOP) policy for a multi-class patient setting is derived through simulation.

Drug Formulary Design Project

Published:

Canada is the 10th largest pharmaceutical market in the world. In 2015, drug sales amounted to $25 billion dollars. Public and private plans constitute 42% and 58% of the drug insurance market respectively. Generics drugs sell at an average of 36% of brand prices, represent 2/3 of prescriptions and amount to 1/4 of total drug expenditures. The goal of this project is to provide a method that helps determine if a new drug should be included in the formulary and whether an existing (i.e., covered) drug should be excluded from the formulary using Multi-Criteria Decision Analysis approach called UTADIS. The method was illustrated using statins data from the National Prescription Drug Utilization Information System (NPDUIS) and applied on oncology drugs using the pan-Canadian Oncology Drug Review (pCODR) recommendations.

House Price Prediction (Regression) Project

Published:

In this project, I’ll develop prediction models using the house prices dataset from Aimes, IA. The goal is to demonstrate the 4 steps of the Data Science project lifecycle: Define, Discover, Develop and Deploy. First, I’ll establish simple baseline model using the OLS regression, and then I’ll develop a few predictive models, namely, random forest, xgboost and lightgbm regression models and compare the performance of these models against the baseline with the aim to get better predictive performance. The implementation of similar price prediction models will potentially allow the housing agencies (e.g., CMHC in Canada), real-estate companies, central and commercial banks, municipial governments and home buyers to make informed decisions with respect to market pricing.GitHub Repo

Loan Eligibility (Imbalanced Classification) Project

Published:

Dream Housing Finance company provides mortgage lending solutions to home buyers. Using this partial dataset, the company wants to automate the loan eligibility process (in real-time) based on customer information upon submission of the online application. These details include Gender, Marital Status, Education, Number of Dependents, Income, Loan Amount, Credit History and others. The goal is to classify the applications into Loan and No Loan classes. To this end, I’ll explore three classification models in this notebook. GitHub Repo

publications

talks

teaching