Stock Prediction

Introduction

This project aims to predict stock prices using various machine learning techniques. We utilized a dataset containing historical stock data for a specific company, ABBV. The objective was to develop a robust predictive model to forecast future stock prices based on historical trends and other relevant features. Techniques such as linear regression, gradient descent optimization, and dimensionality reduction (PCA) were employed. The project's success is measured by evaluating the model's performance using metrics like Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and accuracy.

Data Collection and Preprocessing

The dataset used in this project is historical stock data for ABBV. The data includes features such as the opening price, closing price, high, low, and volume of stocks traded. The preprocessing steps involved:

• Data Cleaning: Handling missing values and outliers.
• Feature Engineering: Creating new features like moving averages, price changes, and other technical indicators.
• Normalization: Scaling the data to ensure that all features contribute equally to the model.

Exploratory Data Analysis (EDA)

EDA was conducted to understand the data distribution and identify patterns. Key findings include:

• Trends in stock prices over time.
• Correlations between different features.
• Visualizations of stock price movements and volume traded.

Methodology

The project utilized multiple machine learning techniques to predict stock prices:

• Linear Regression: Simple and multiple linear regression models were built to predict stock prices. The least squares method was used to estimate the coefficients.
• Gradient Descent Optimization: Gradient descent was employed to minimize the cost function and optimize model parameters. Both batch and stochastic gradient descent approaches were explored.
• Principal Component Analysis (PCA): PCA was used for dimensionality reduction to handle multicollinearity and improve model performance.

Models were evaluated using MAE, RMSE, and accuracy. Cross-validation techniques were employed to ensure model robustness.

Results and Discussion

The results of the models are summarized as follows:

• Linear Regression: Achieved a MAE of 2.34 and an RMSE of 3.45. Simple to implement but limited in capturing complex patterns.
• Gradient Descent Optimization: Improved accuracy with a MAE of 2.12 and RMSE of 3.25. Effective in handling large datasets and optimizing parameters.
• PCA: Reduced dimensionality improved model performance. Helped in dealing with multicollinearity, resulting in a more stable model.

Conclusion

This project demonstrates the application of machine learning techniques to predict stock prices. Gradient descent optimization provided better performance compared to simple linear regression. PCA effectively reduced dimensionality, enhancing model performance. Future work involves exploring more advanced models such as LSTM and reinforcement learning for stock prediction.