📊 NVIDIA Stock Price Prediction Pipeline

This repository contains a robust, end-to-end pipeline for predicting the short- to medium-term (1 week to 1 month) price movement of NVIDIA (NVDA) stock using technical indicators and macroeconomic variables.

---

🚀 Project Overview

This project combines data preparation, feature engineering, target generation, and a suite of regression and classification models. It evaluates models using both traditional accuracy metrics and financial strategy metrics to determine their practical profitability.

---

🧱 Pipeline Structure

The pipeline is modular and controlled through a single master script: data_pipeline.R.

🗂️ Modules Included

• Data Clean

  • import_data.R: Pulls stock and macroeconomic data.
  • handle_missing.R: Cleans and fills missing values using time-series forward fill.

• Data preparation

  • feature_engineering.R: Builds technical and macro features.
  • clean_column_structure.R: Cleans and aligns column names and structures.
  • data_quality_checks.R: Checks consistency, types, NAs, etc.
  • add_targets.R: Generates price and direction targets.

• Exploratory Data Analysis

• univariate_distribution.R: Plots histograms and density plots of technical indicators (RSI, MACD, Volatility, Volume, Return).

• Time-series_plots.R: Generates time series plots for key indicators (Close price, RSI, MACD, Volatility).

• Stationarity_Test.R: Performs ADF and KPSS tests on log returns; saves results and plots.

• ACF_PACF_plots.R: Plots ACF and PACF of log returns to explore autocorrelation structure.

• Scatter_plots.R: Shows scatter plots of each feature against target variables to assess linear relationships.

• feature_response_relationship.R: Visualizes how selected features relate to price or directional targets using smoothed plots.

• technical_indicator_boxplots.R: Creates boxplots grouped by direction target to compare feature distributions across classes.

• Correlation_Heatmap.R: Displays a correlation heatmap across all numeric predictors and targets.

• Modeling Setup

  • 00_modeling_setup.R: Defines feature sets, target variables, and output folders.
  • 01_data_split.R: Performs time-based split into train/test sets.
  • 03_modeling_utils.R: Evaluation and plotting functions for regression models.
  • 04_strategy_evaluation_utils.R: Trading strategy metrics (returns, Sharpe, drawdown).

• Models Implemented

  • Baseline
  • Linear Regression
  • Lasso Regression
  • Ridge Regression
  • XGBoost Regressor
  • XGBoost Classifier (standard + tuned)
  • Random Forest
  • Support Vector Regression (SVR)
  • Random Forest Tuned

---

▶️ Running the Pipeline

To run the full pipeline from start to finish, including data loading, cleaning, modeling, and evaluation:

source("data_pipeline.R")

You can control which models to run by toggling the following flags inside data_pipeline.R:

run_linear <- TRUE
run_lasso <- TRUE
run_ridge <- TRUE
run_xgboost <- TRUE
run_xgboost_tuning <- TRUE

🧮 Model Evaluation

Each model is evaluated using:

📉 Regression Metrics

• RMSE (Root Mean Squared Error)
• MAE (Mean Absolute Error)
• MAPE (Mean Absolute Percentage Error)
• R² (Coefficient of Determination)

💰 Strategy Metrics

• Cumulative Return: Total profit/loss from trading
• Sharpe Ratio: Risk-adjusted return (target ≥ 0.8)
• Max Drawdown: Worst loss from peak (target ≤ 15%)
• Directional Accuracy: Percent of times model predicted the right direction

---

📁 Output Structure

All outputs are stored under the /Results/ directory:

🔹 Model-specific folders:

• /Results/LinearRegression_*
• /Results/XGBoost_Regression_*
• /Results/XGBoost_Classifier_Tuned_*

📄 Files inside each model folder:

• *_predictions.csv: Actual vs. predicted values
• *_plot.png: Time-series plot of predictions

📊 Centralized log:

• Results/all_model_metrics.csv: Combined metrics across all models

---

📌 Technologies Used

• Language: R
• Libraries:
  xgboost, caret, dplyr, quantmod, TTR, zoo, ggplot2, scales
• Data Sources:
  • Yahoo Finance (via quantmod)
  • FRED (Federal Reserve Economic Data)