amitbitan/gold-price-prediction-amit

Gold Price Prediction Analysis (1970-2024)

Project Overview This project performs a comprehensive analysis and price prediction of Gold using historical market data. The workflow includes data cleaning,EDA, and building a predictive model using Linear Regression.

Research Questions & Answers

1. Price Distribution

• Question: How are gold prices distributed over the decades? Are there specific frequent price ranges?
• Answer:The distribution is bimodal, showing that gold prices often consolidate around specific psychological levels (like $1200) before breaking out to new historical highs.

2. Feature Correlation

• Question: Which market indicators (Open, High, Low, Volume) most strongly impact the closing price?
• Answer: There is a near perfect correlation (1.00) between intraday prices (High/Low) and the closing price, suggesting that daily volatility is highly consistent.

Key Visualizations & Insights

• Historical Trend: We visualized the overall growth of gold, identifying massive spikes during global economic shifts.
• Correlation Heatmap: This map confirmed that features like Open and High are the strongest predictors for our model.
• Actual vs. Predicted Plot: Our final visualization shows the model's predictions (blue dots) sitting almost perfectly on the actual price line (red dashed line).

Decisions & Data Cleaning

• Missing Values: I identified and removed null rows using dropna() to ensure model stability.
• Feature Engineering: I created a new feature, Daily_Range (High - Low), to capture market volatility.
• Outlier Decision: I analyzed price outliers using a Boxplot. I decided to keep these outliers because they represent real market spikes rather than data entry errors.

Model Performance (End Result)

• Algorithm: Linear Regression
• R² Score:0.99963
• Conclusion: The model is exceptionally accurate at predicting the daily closing price based on intraday features.

Author: Amit Bitan
Tools used: Python, Pandas, Scikit-Learn, Seaborn, Matplotlib. license: mit