Title: Time Series Forecasting Project
Description:
This GitHub repository showcases a comprehensive project undertaken for the "Streaming Data Management and Time Series Analysis" examination. The primary objective of this project was to provide accurate forecasts for the time period spanning from April 1, 2015, to November 7, 2015. The forecasting task was approached using three distinct methodologies: an ARIMA model, a UCM (Unobserved Components Model), and a machine learning-based model.
The project workflow involved a rigorous validation process to evaluate and compare the performance of the forecasting algorithms. The goal was to identify the most promising models within each family (ARIMA, UCM, and Machine Learning). Model evaluation was conducted based on the Mean Absolute Error (MAE) metric, a mathematical measure representing the average absolute difference between the predicted and actual values. The lower the MAE, the better the model's predictive accuracy.
Key Project Components:
-
ARIMA Model Implementation:
- Utilized the AutoRegressive Integrated Moving Average (ARIMA) model to capture time-series patterns and make predictions for the specified timeframe.
-
UCM Model Implementation:
- Implemented an Unobserved Components Model (UCM), leveraging its ability to decompose time series into various components (trend, seasonality, etc.) for improved forecasting.
-
Machine Learning Model Implementation:
- Employed a machine learning-based forecasting approach, exploring various algorithms to identify the most effective model for the given time series data.
-
Validation and Model Selection:
- Executed thorough validation processes to assess the performance of each model.
- Selected the top-performing algorithms from each family based on their MAE scores, ultimately narrowing down the choices to the three "best" models.
This repository serves as a valuable resource for anyone interested in time series forecasting, providing detailed insights into the implementation of ARIMA, UCM, and machine learning models, along with the evaluation criteria employed to identify the most accurate forecasting methods.