Masters Student

Posts by Collection

portfolio

Simulation Study: Hyperparameter Tuning with the DoubleML Package

This is a project I am currently working on for the chair of statistics at the business faculty of the University of Hamburg. It seeks to create a programming framework in python to train multiple high-dimensional data sets using different parameter tuning strategies as well as different splits with the DoubleML package for causal machine learning. The parametric partially linear regression model is applied and performance measures such as bias, coverage and mean squared error are recorded for the different training settings. Learners such as lasso, random forest, gradient boosting for regression and for classification are applied to estimating the nuisance functions and a combination of the best learners is chosen. Results are presented on per learner and per strategy basis.

Statistical Programing with Python

This project is the application of different statistical and machine learning methods as well as simulations and graphics all conducted with python. It includes basic statistical methods such confidence intervals, hypothesis testing, statistical distributions as well as empirically showing the statistical rules, such as the law of large numbers.

Neural Network Evaluation

This evaluation notebook writen by me on Kaggle was used to evaluate the different neural networks trained in the classification of NIH-X-Ray Images for a project with the MIN Faculty at the University of Hamburg.

Thesis: Demand Estimation with Machine Learning

This paper describes a comparison of classical and machine learning methods for the prediction of demand of breakfast cereals. A literature review on predicting demand with discrete choice models and machine learning methods is provided. A utility based linear demand model serves as theoretical basis. The applied methods are explained, then data is simulated to allow for the assignment of coefficients and variation of features. Data pre-processing is done before fitting the models. Results show that machine learning methods outperform classical methods, when working with large number of features.

Natural Language Processing with Python: Amazon Reviews

This project involves the loading and pre-processing of amazon reviews and their ratings, the fitting of an artificial Deep Neural Network capable of being trained on up to 3,6 million observations (the size of the sample). A smaller sample is used for computational speed, however the methods don’t change. Implementation is done using Python and a remote GPU.

Predicting Survival Rates on the Titanic

In this project the Titanic survivor data set is used. The goal is to fit a model that accurately predicts which passengers survived by learning the relationships between the individual passenger-characteristics and their survival rate. The data set is first studied to determine what pre-processing procedures or ML methods are necessary. A simple model with the original features and a complex model with engineered features is applied to test for interactions or non-linear relationships between the variables. Data is split into training and testing using cross-validation-resampling. The models are tuned using grid search and performance is measured using accuracy score. An ensemble model combines all the models into one. A prediction with the ensemble model is submitted to Kaggle.com using the provided hold out set.

Marketing Seminar: Behavioral Experiment with Product Shortages

This was a behavioral decision-making experiment designed and conducted by me as part of a marketing seminar at the University of Hamburg. The goal was to show that consumers will adjust their demand for products that are uncertain, in this case the availability of car sharing nearby. A survey was designed that placed the subjects in a demand situation for car sharing with one group receiving a product that was heavily demanded, while the other was the control group. The product shortage was perceived by the subjects, who were recruited both on social media as well as on the platform Amazon Mechanical Turk. An appropriate survey with attention-checks and an optional lottery was designed to encourage participation. The collected data was checked for accuracy and evaluated using SPSS.

Corporate Governance: Discussing the Merits and Demerits of Stock Options

The purpose of including stock options and restricted stock in a contract is to align management’s incentives with the incentives of the shareholders of the company and to also mitigate the agency problem. The agency problem is caused by managers who act in their best interests instead of the interests of the owners of the company. Therefore stock options and restricted stock is issued as part of compensation packages of managers to incentivize them to work in the best interest of the company. If the company stock price raises their options and restricted stock will appreciate as well therefore aligning their interests with the rest of the shareholders.

publications

talks

teaching

Teaching experience 1

Undergraduate course, University 1, Department, 2014

This is a description of a teaching experience. You can use markdown like any other post.

Teaching experience 2

Workshop, University 1, Department, 2015

This is a description of a teaching experience. You can use markdown like any other post.