Aleksei Luchinsky

Students Performance Analysis

​In this report, I analyzed student's performance dataset and tried to find out, why students do fail their classes. Both logistic regression and linear regression models are presented. The first one makes the classification, trying to predict, whether the student will fail his/her class or not. The second, linear regression, model predicts final grades of the students, that did not fail. Both models show good prediction accuracy both on training and test data sets. Some discussion about significant factors can be found at the end of the report.

Machine Learning in Board Games

Unsupervised machine learning programs for playing some simple board games are considered. For the simplest one, naught and crosses game on a 3 × 3 board the program has found an optimal strategy without any prior knowledge about the game. In the case of a more complicated game on 5 × 5 playing board, the program sometimes makes errors, missing the evident winning moves.

Parallel Calculation of the Eigenvalues

This is a final project for Dr. Green’s CS5170 class

Keywords Analysis in High Energy Physics Publications

(with Jacob Mitchell, Xuejun Gu)

Clusterization of published articles in high energy physics based on keywords, extracted automatically from titles and abstracts of the papers

SAI Dashboard

(Systems Associates Inc. - SAI)

Online system for creating, modifying, and using dashboards that helps to get KPI information about the hotel performance

Function Data Analysis of the Stock Market Prices Since the Pandemic

(MS Project)

The subject of the project is the analysis of available Stock Market Prices data since the pandemic using the Functional Data Analysis (FDA) approach. The data were cleaned and transformed before the analysis using FDA. We applied functional principal component analysis (FPCA) and functional clustering (FC) to the data. Using FDA, the companies were clustered into an optimal number of groups required to explain the observed variation, after that the analysis and comparison of the clusters’ content was performed. Our results show that during the pandemic period (from March 13,2020 to the exact date) mostly increasing trends were observed, but the increasing trends are different from one cluster to another. Interesting findings on the industry sections and obtained clusters are also elaborated based on FC results.

Factors Correlated with Life Expectancy Around the World

(with Jingyi Su, Kim Brooks, Vibhuti Chandna)

(GitLab)

The paper is devoted to inspect the World Health Organization data on the life expectancy and to find out what factors correlate with the life expectancy the most. AThe final linear regression model explains 80% of the variance using only 6 regressor variables (including alcohol consumption, government expenditure on health, adult Mortality Rates, HIV/AIDS death rate, country development status, and interactions).

Jarque –Bera Test and its Competitors for Testing Normality – A Power Comparison

(with Jishan Ahmed, Donghyun Jeon, Upeksha Perera)

(GitLab)

In this paper, we attempted to reproduce the results to validate the claims which were made by authors that the power of Jarque –Bera test is better for the symmetric distribution, whereas it performs poorly on the bimodal distribution.

Multiple Sclerosis Statistics Analysis

(with Kim Brooks, Vibhuti Chandna, Dong Hyun Jeon)

(GitLab)

This paper is devoted to the analysis of Multiple Sclerosis pattern in the United States. To perform the analysis, the data provided by Medical Expenditure Panel Survey was used. Methods such as Decision Tree, Logistic Regression, Neural Network, k-Nearest Neighbors, Random Forest, Adaptive Boosting, and Linear Regression were implemented to predict the probability of a person being diagnosed with MS and determine which demographic factors are important for answering this question.

COVID-19: Death Statistical Analysis

(with Michael Terry and Vagish Vela)

(GitLab)

In this report, we study the impact of COVID- 19 on several US states and how the disease compares with historical causes of death, and it’s potential contribution to excess deaths when compared to an equivalent historical time frame. We provide insights into how our data can support excess death analysis in the future, and reflect on the significant challenges, highlighted by our study, for future research in comparing COVID-19 datasets.