Univariate Time Series Forecasting

This project provides implementations for univariate time series forecasting using ARIMA, Auto-ARIMA, and LSTM models. The code is modular and designed for easy adaptation to various datasets and forecasting requirements.

Table of Contents

1. Introduction
2. Features
3. Installation
4. Usage
5. Data Preparation
6. Models
7. Evaluation
8. Results
9. License

Introduction

This repository contains code for forecasting univariate time series data using ARIMA, Auto-ARIMA, and LSTM models. The project is structured to be easily extensible and reusable for different datasets and forecasting scenarios.

Features

• Data preparation and scaling
• Hyperparameter tuning for ARIMA and LSTM models
• Rolling and expanding window evaluation methods
• Plotting results and summary metrics

Installation

Prerequisites

• Python 3.7 or later
• Required packages listed in requirements.txt

Installing

Clone the repository:

git clone https://github.com/zhaleh-rahimi/univariate-time-series-forecasting.git
cd univariate-time-series-forecasting

Install the required packages:

pip install -r requirements.txt

Usage

Command Line Interface

Run the main script with the necessary arguments:

python main.py --file_path data/your_data.csv --date_col date --target_col target --model_type ARIMA --evaluation_type rolling --output_file results/output.csv --steps 4

Arguments

• --file_path: Path to the CSV file containing the time series data
• --date_col: Name of the date column in the CSV file
• --target_col: Name of the target column in the CSV file
• --model_type: Type of model to use (ARIMA, AUTO-ARIMA, LSTM)
• --evaluation_type: Type of evaluation method (rolling, expanding)
• --output_file: Path to save the output CSV file
• --steps: Number of steps ahead to forecast

Data Preparation

Loading Data

The data is loaded using the DataPrep class, which also handles scaling and splitting the data into training and testing sets.It also can prepare a dataset for a supervised learning process, e.g. LSTM forecasting.

Scaling Data

Data is scaled using the MinMaxScaler from sklearn.preprocessing. One may change the scaler.

Models

ARIMA

Implemented in the ARIMAModel class. Hyperparameters are selected using grid search and minimizing mse for imeplemented ARIMA.

Auto-ARIMA

Implemented in the AutoARIMA class using the pmdarima library.

LSTM

Implemented in the LSTMModel class. Hyperparameters are selected using grid search.

Evaluation

Rolling Window Evaluation

Evaluates the model performance using a rolling window approach.

Expanding Window Evaluation

Evaluates the model performance using an expanding window approach.

Results

The results are saved to a CSV file and can be visualized using the provided plotting functions.

License

This project is licensed under the MIT License - see the LICENSE file for details.