Novo Nordisk
Data Science Intern / Co-op | Commercial Insights & Analytics | Impact Analysis & Market Investment
Novo Nordisk Overview
Novo Nordisk is a renowned global healthcare company specializing in diabetes care and other serious chronic conditions. Founded in 1923 in Denmark, the company has since become a leading player in the pharmaceutical industry, with its headquarters located in Bagsværd, Denmark.At the core of Novo Nordisk's mission is its dedication to improving the lives of people living with diabetes and other chronic diseases. The company is a pioneer in the field of diabetes treatment, offering a wide range of innovative medications, devices, and therapies to manage the condition effectively.
Skills / Software used
Software/platforms/Language: Python, Pyspark, SQL, Pytorch, Databrick, Dataiku, Snowflake, Tableau
Knowledge/Skills: Machine Learning, Data Analytics, Causal Inference, Applied Linear Model, Time Series
Knowledge/Skills: Machine Learning, Data Analytics, Causal Inference, Applied Linear Model, Time Series
My Contributions
I work for the function of Commercial Insights & Analytics function, our function's goal is to use data modeling to support campaign activities and conduct market research.
During the internship, I have finished two individual project and participated to the other project - I made excellent performance that I received the offer for extension.
The first project I finished is the Test-Control Impact Analysis back by Machine Learning (Random Forest) based Propensity Score Matching Techniques. I built the feature store and the model by myself and I have introduced Bartacharyya Coefficient and Standardized Mean Difference as the measurement for "matching performance", and I also created the dashboard to visualize the sales lift (post-period) and presented during the marketing readout.
Moreover, I have fully automated the ML pipeline to make the model reusable.
The second project is the Linear Model based Marketing Mix analysis, which use linear model to measure the degree of impact for different marketing campaigns' effect on the post period sales performance. I achieved the goal to bring this model in-house and establish the standard for the future marketing mix model's reference.
I also successfully addressed granularity problem, and the problem of feature engineering stage.
The last project I participated is the Non-Personal Promotion target project based on Machine Learning, which aims to use ML algorithm to predict the HCP that is suitable for the next period non-personal promotion. I participated in renewing the feature store and increase the model (LightGBM) robustness and stability for different backtesting period, and I revised this model to overcome the input features' trend/seasonality problem by introducing features representing lag, difference, and moving average.
During the internship, I have finished two individual project and participated to the other project - I made excellent performance that I received the offer for extension.
The first project I finished is the Test-Control Impact Analysis back by Machine Learning (Random Forest) based Propensity Score Matching Techniques. I built the feature store and the model by myself and I have introduced Bartacharyya Coefficient and Standardized Mean Difference as the measurement for "matching performance", and I also created the dashboard to visualize the sales lift (post-period) and presented during the marketing readout.
Moreover, I have fully automated the ML pipeline to make the model reusable.
The second project is the Linear Model based Marketing Mix analysis, which use linear model to measure the degree of impact for different marketing campaigns' effect on the post period sales performance. I achieved the goal to bring this model in-house and establish the standard for the future marketing mix model's reference.
I also successfully addressed granularity problem, and the problem of feature engineering stage.
The last project I participated is the Non-Personal Promotion target project based on Machine Learning, which aims to use ML algorithm to predict the HCP that is suitable for the next period non-personal promotion. I participated in renewing the feature store and increase the model (LightGBM) robustness and stability for different backtesting period, and I revised this model to overcome the input features' trend/seasonality problem by introducing features representing lag, difference, and moving average.
