Leveraging Machine Learning for Time Series Predictive Analysis

This GitHub repository is a code base for a Diploma thesis.

Installation

For running jupyter notebooks in the project (files which end with .ipynb), you will need to get the interface (https://docs.jupyter.org/en/latest/install.html). Basics for running notebook: https://unidata.github.io/python-training/python/notebook/.

The requirements for this repository can be dowloaded with:

pip install -r requirements.txt

Time Series Prediction Script (tool.py)

Overview

This Python script is capable of predicting data up to 30 days/hours (depending on data set). The supported data sets are also included with whole research in their related folder

Usage

To use this script, you will need to specify the model and dataset when running the script from the command line. Here is a basic example of how to invoke the script:

python timeseries_prediction.py --model lstm --dataset btc --forecastLength 15

Parameters

• --model (-m): The prediction model to use. Supported options include regression, arima, sarima, lstm, prophet, prophet_log, transformer, dlt, ets, ktr, automl, gnn.
• --dataset (-d): The dataset to perform predictions on. Currently supports covid_deaths and btc.
• --forecastLength (-fc): The length of the forecast, with a maximum value of 30 due to model limitations.

Data Sources

The script fetches data from publicly available APIs:

• COVID-19 data from the World Health Organization.
• Bitcoin price data from Yahoo Finance.
• Electricity demand from https://www.eia.gov/

Ensure network access is available to fetch the latest data for predictions.

Customizing the Script

The script is modular, allowing for easy expansion to include more datasets or models. To add a new dataset, define its structure in dataset_metadata and implement corresponding loading and preprocessing functions.