An Analysis of Predictive Models for Forecasting 2010-2022 TESLA INC. Stock Price

ENG178 Statistics and Data Science for Engineers course final project.

Project Goal: To develop a model that predicts daily closing price evolution over time of Tesla’s stock price using a dataset consisting of opening and closing stock price of TSLA between 2010-2022.

Modeling Process:
- Evaluating raw dataset statistics
- Data pre-processing: normalization, baseline, and evaluation helper functions
- Chunking the dataset between training, validation, and test sub-datasets
- Evaluating the time-series performance of various models, including linear regression, neural networks (MLP, SimpleRNN, LSTM), and a non-linear, tree-based model, Random Forests.
- Adjusting hyperparameters (training viewing window size) of best-performing model (linear regression)

We concluded that linear regression was best for predicting closing price of this dataset based on Train Margin Mean Absolute Error (MAE), Validation MAE, and Test MAE, as well as training time (in seconds).

My responsibilities within group project:
- Modeling linear regression and iterating through hyperparameter adjustments
- Modeling Random forests
- Modeling MLP, SimpleRNN neural networks
- Raw dataset statistical analysis
- Hosting and scheduling team meetings and delegating deliverables to meet deadlines
- Writing and formatting final written report deliverable in LaTeX.

Technologies Used: Python, Jupyter Notebook, Visual Studio Code, LaTeX

Future additions: Although external factors such as quarterly
earnings, car charging (supercharger) prices, major news, or CEO Elon Musk’s public actions (such as tweeting on X) can strongly influence Tesla’s stock price, our initial focus is on the TSLA stock dataset itself. External information like these could later be added as additional features.