Michael Son

Michael Son

Research Scientist | Data Scientist Data Analyst | ML Engineer

Professional Summary

Meticulous and analytical researcher with expertise in molecular biology techniques and data analysis. Detail-oriented and methodical professional proficient in conducting rigorous experimental research utilizing state-of-the-art equipment. Familiar with gathering, cleaning, and organizing data for use by technical and non-technical personnel. Advanced understanding of statistical, algebraic, and other analytical techniques.

Download my résumé.

Interests
  • Biotechnology
  • Data Science
  • Machine Learning
Education

  • MSc in Chemical and Bioengineering, 2019

    Friedrich-Alexander-Universität Erlangen-Nürnberg

  • HBSc in Neuroscience/Cell Molecular Bio, 2014

    University of Toronto - St. George

Skills

Coding
Data Science
Machine Learning
Molecular Biology Techniques
Experimental Research

Projects

*
All Supervised Learning Unsupervised Learning Classification Regression
Lending Club Loan Approval

Summary

  • Lending Club 2007-2011 dataset with 42,538 entries and 115 features.

  • Tools Used: NumPy | pandas | Matplotlib | Seaborn | Scikit-Learn | XGBoost | Imbalanced-Learn

  • Constructed a binary classifier based on financial attributes of loan applicants to differentiate unprofitable loans from the rest.

    • 1: Good/Profitable Loan | Majority Class
    • 0: Bad/Unprofitable Loan | Minority Class

  • Addressed the issue of imbalanced data (6:1 class ratio) by oversampling the minority class using SMOTE from Imbalanced-Learn.

  • Models Trained/Evaluated: Logistic Regression | Random Forest | Support Vector Machine | eXtreme Gradient Boosting

  • Optimized models by tuning their hyperparameters with RandomizedSearchCV.

  • Best Model: eXtreme Gradient Boosting

  • Best Model Performance: Minority F1 = 0.56 | Majority F1 = 0.89 | ROC AUC = 0.80 | Accuracy = 83%

Follow
Melanoma Tumor Size Prediction

Summary

  • Datasets provided by Machine Hack: Melanoma Tumor Size Prediction

    • Training Set: 9,146 entries, 10 features
    • Test Set: 36,584 entries, 10 features

  • Tools Used: NumPy | pandas | Matplotlib | Seaborn | Scikit-Learn | Keras

  • Constructed a regressor to predict the numerical value of melanoma tumor size based on relevant attributes.

  • Models Trained/Evaluated: Multiple Linear Regression | Random Forest | Support Vector Machine | Multi-Layer Perceptron | Keras Regression

  • Optimized models by tuning their hyperparameters with RandomizedSearchCV.

  • Best Model: Random Forest

  • Best Model Performance: MSE = 8.25 | R2 = 0.23

Follow

Contact