Economics & Machine Learning
A collection of applied machine learning projects using financial, music, and business datasets. These projects focus on prediction, model comparison, feature engineering, and interpreting results in an economic or business context.
Projects
Predicting Stock Market Direction
R · Logistic Regression · LDA · QDA · kNN · ROC-AUC
Tested whether lagged S&P 500 returns could predict daily market direction using linear and nonlinear classification models.
Highlights
- Built logistic regression, LDA, QDA, and kNN models
- Used time-based train/test splitting
- Evaluated cross-validation, confusion matrices, ROC curves, and AUC
- Found limited predictive power, consistent with weak-form market efficiency
Predicting Spotify Song Popularity
R · Regression · LASSO · Random Forest · XGBoost · GAMs
Compared linear and nonlinear machine learning models to predict Spotify song popularity using audio features and track metadata.
Highlights
- Cleaned and modeled 100K+ Spotify tracks
- Compared regression, Ridge, LASSO, Elastic Net, GAMs, Random Forest, Bagging, and XGBoost
- Evaluated RMSE, MAE, and R²
- Used bootstrap confidence intervals for model performance
Predicting Restaurant Success
R · Classification · Random Forest · Boosting · SVM · SMOTE
Built classification models to predict restaurant success using marketing, customer demand, operations, and review-based variables.
Highlights
- Compared classification trees, Random Forest, Boosting, Neural Networks, and SVM
- Used SMOTE to address class imbalance
- Evaluated accuracy, balanced accuracy, ROC curves, and ROC-AUC
- Identified marketing budget, social media followers, Google ratings, and reservations as key predictors