Stock Price Prediction

Objective

The primary objective of this project is to develop a machine learning model capable of forecasting the future stock prices of Microsoft Corporation using historical stock price data. By implementing Long Short-Term Memory (LSTM) neural networks, we aim to leverage advanced deep learning techniques to provide accurate predictions of Microsoft's stock prices.

Introduction

The ability to predict stock prices accurately has always been a challenging task in the financial world. The volatility and complexity of financial markets make it a perfect candidate for advanced machine learning techniques. In this project, I embarked on the task of forecasting Microsoft Corporation's stock price using Long Short-Term Memory (LSTM) neural networks.

This project holds significant importance not only as a practical application but also as a valuable addition to my resume, showcasing my skills in data analysis, machine learning, and financial forecasting.

Data Collection and Preprocessing

To begin the project, I first obtained the historical stock price data for Microsoft Corporation from Yahoo Finance. The dataset spans from the company's inception in 1986 to the present day (or the most recent data available). The data included various attributes, such as date, open price, high price, low price, closing price, and trading volume.

For simplicity, I decided to focus solely on the closing price as the target variable for forecasting. I extracted and preprocessed the relevant data, converting the date column into proper datetime objects and creating a DataFrame containing only date and closing price columns.

Data Transformation for LSTM

To apply LSTM neural networks for stock price forecasting, I needed to reformat the dataset into a supervised learning problem. This involved creating a new DataFrame where each row represents a target date and includes the closing prices from the three previous days. This arrangement allows the model to predict the next day's closing price based on historical data.

I provided code to demonstrate how this data transformation was performed, creating the necessary input sequences and corresponding target values.

Model Development

For the modeling phase, I utilized TensorFlow and Keras to build an LSTM-based neural network. The architecture of the model consisted of an input layer, an LSTM layer with 64 neurons, two dense layers with 32 neurons each, and a final output layer with a linear activation function.

The model was compiled with mean squared error (MSE) as the loss function and the Adam optimizer with a learning rate of 0.001. Additionally, I used mean absolute error (MAE) as a metric to evaluate the model's performance.

Model Training and Evaluation

The dataset was split into training, validation, and testing sets to train and evaluate the model. Training was performed for 100 epochs, and the training process was visualized using loss curves. While the training data exhibited good performance, the validation and testing sets provided valuable insights into the model's generalization capabilities.

I visualized the model's predictions against actual stock prices for the training, validation, and testing datasets. This visualization helped demonstrate how well the model performed, with a particular focus on its ability to predict future stock prices.

Recursive Prediction Experiment

In an attempt to predict stock prices beyond the initial dataset, I conducted a recursive prediction experiment. This involved using the model to predict stock prices for future dates iteratively, feeding the predicted values back into the model as input.

The results of this experiment highlighted the model's limitations in extrapolating beyond the training data. It became evident that predicting stock prices over extended periods is a complex task due to the inherent unpredictability and ever-changing nature of financial markets.

Conclusion

This project provided an opportunity to apply LSTM neural networks to a real-world financial forecasting problem. While the model demonstrated promising results within the observed data range, it faced significant challenges when attempting to predict stock prices beyond that range.

The project serves as a valuable addition to my portfolio, showcasing my skills in data preprocessing, model development, and data visualization. It also highlights the importance of understanding the limitations of machine learning models, especially in the context of financial forecasting.