📈 Quantageddon 1.0

A Time-Series Feature Engineering and Ridge Regression Project

---

🎯 Objective

Quantageddon 1.0 is a machine learning project designed to explore feature engineering and regression modeling for time-series financial data. The focus is on predicting the target variable "Open" (possibly stock prices) by generating meaningful features and applying a Ridge regression model with hyperparameter tuning.

---

🧪 Pipeline Overview

📊 Phase 1: Feature Engineering

• Generated time-based features from datetime columns (e.g., day of week, day of year).
• Created interaction features by combining original variables.
• Computed Pearson correlations between features and target (Open) to evaluate usefulness.
• Removed features with low or negative correlation to reduce noise and overfitting.

🛠️ Tools Used:

• pandas, numpy for data handling
• seaborn, matplotlib for visualization
• sklearn for modeling and evaluation

---

🤖 Phase 2: Modeling with Ridge Regression

• Used Ridge Regression (L2-regularized linear model) from sklearn.linear_model
• Implemented GridSearchCV for hyperparameter tuning of the alpha parameter
• Trained the model on engineered features
• Evaluated performance using standard regression metrics
• Final predictions were exported as a .csv file for submission or further use

---

📤 Output

• submission.csv: contains predicted Open values on the test set
• Feature importance & correlation plots for interpretability

---

📁 Project Structure

• quantageddon-1-0.ipynb: Jupyter notebook with full code and markdown explanations
• submission.csv: Final model predictions
• Plots and metrics embedded in notebook

---

🧠 Key Takeaways

• Feature engineering significantly boosted model performance
• Time-series context was captured using periodic features
• Ridge Regression provided a stable and interpretable baseline
• GridSearchCV helped fine-tune model parameters efficiently

---

🚀 Future Scope

• Explore other regularized models (Lasso, ElasticNet)
• Add lag-based features or rolling window statistics
• Use tree-based models (e.g., XGBoost, LightGBM) for comparison
• Extend to multi-target prediction (e.g., Open, Close, High, Low)