Paper: arXiv 2405.04539

Authors: Aryan Bhambu, Arabin Kumar Dey

Abstract

This research paper introduces innovative approaches for multivariate time series forecasting based on different variations of the combined regression strategy. We use specific data preprocessing techniques which makes a radical change in the behaviour of prediction. We compare the performance of the model based on two types of hyper-parameter tuning Bayesian optimisation (BO) and Usual Grid search. Our proposed methodologies outperform all state-of-the-art comparative models. We illustrate the methodologies through eight time series datasets from three categories: cryptocurrency, stock index, and short-term load forecasting.

Complexity vs Empirical Score

  • Math Complexity: 7.0/10
  • Empirical Rigor: 8.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper introduces a novel ensemble method with advanced statistical formulations and detailed algorithmic descriptions, but also validates its claims with multiple real-world datasets, hyperparameter tuning comparisons, and clear performance metrics.

Research Flowchart

  flowchart TD
  A["Research Goal: Optimize Multivariate Time Series Forecasting"] --> B["Data Collection & Preprocessing"]
  B --> C["Three Datasets:<br>Stock Index, Crypto, Load Forecasting"]
  C --> D["Model Architecture:<br>COBRA Variations"]
  D --> E["Hyper-parameter Tuning:<br>BO vs. Grid Search"]
  E --> F["Computational Process:<br>Training & Forecasting"]
  F --> G["Key Findings:<br>1. Proposed methods outperform SOTA<br>2. BO > Grid Search<br>3. Preprocessing crucial"]

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper proposes two new COBRA variations, Dynamic Proximity Ensemble and Partitioned Dynamic Proximity Ensemble, for multivariate time‑series forecasting.

Source sentence
We explore two different variation of COBRA, Dynamic Proximity Ensemble and Partitioned Dynamic Proximity Ensemble respectively for multivariate time series forecasting.

Assumes 4

  • The highest and lowest values within the training set are represented by xmax and xmin for max‑min normalisation. data
    Source sentence
    We assume that the highest and lowest values within the training set represented by xmax and xmin, respectively.
  • The validation and test sets each contain 10% of the entire dataset. data
    Source sentence
    We choose the validation and test sets each with 10% of the entire dataset.
  • COBRA in its current form is valid only for independent datasets (i.e., usual regression settings). model
    Source sentence
    The current form of COBRA available in the literature so far is valid for independent datasets which is meant for usual regression purpose
  • A sliding window creates exchangeable sets so that time‑series can be treated like independent regression samples. model
    Source sentence
    We use sliding window to create the frames. It will help to transfer the whole data into series of exchangeable datasets with a specific window length of covariates and a target variable to perform COBRA on these exchangeable sets.

Data

  • Bitcoin · Yahoo Finance crypto-onchain public
    2015-01-01 to 2023-06-30
    Source sentence
    The Bitcoin dataset represents the cryptocurrency’s daily market performance.
  • DJI · Yahoo Finance market-prices public
    2015-01-01 to 2023-06-30
    Source sentence
    The DJI dataset tracks the Dow-Jones Industrial Average, reflecting the performance of 30 major U.S. companies.
  • S&P500 · Yahoo Finance market-prices public
    2015-01-01 to 2023-06-30
    Source sentence
    The S&P500 dataset covers 500 large-cap American companies, providing a comprehensive view of the U.S. stock market.
  • South Australia (SA) load · Australian Energy Market Operator (AEMO) other public
    December 2022
    Source sentence
    The subsequent five datasets encompass load data collected from the states of South Australia (SA), New South Wales (NSW), Victoria (VIC), and Tasmania (TAS) for the month of December 2022.
  • New South Wales (NSW) load · Australian Energy Market Operator (AEMO) other public
    December 2022
    Source sentence
    The subsequent five datasets encompass load data collected from the states of South Australia (SA), New South Wales (NSW), Victoria (VIC), and Tasmania (TAS) for the month of December 2022.
  • Victoria (VIC) load · Australian Energy Market Operator (AEMO) other public
    December 2022
    Source sentence
    The subsequent five datasets encompass load data collected from the states of South Australia (SA), New South Wales (NSW), Victoria (VIC), and Tasmania (TAS) for the month of December 2022.

Recognised sources: Yahoo Finance

Costs

Not applicable The paper reports no trading or execution result.

Finds 4

  • Dynamic Proximity Ensemble outperforms all other COBRA variations and existing state‑of‑the‑art models on the two evaluation metrics. performance
    Source sentence
    We observe Dynamic Proximity Ensemble outperforms all other variations of COBRA and existing state or the models with respect to the two metrics considered in this paper.
  • BOADPE (Bayesian‑optimised DPE) outperforms all models in both RMSE and MAPE. performance
    Source sentence
    Moreover, BOADPE outperforms all models in both the metrics considered in this paper.
  • BOA‑based models consistently outperform the grid‑search based counterparts, highlighting the robustness of the Bayesian optimisation approach. performance
    Source sentence
    The above comparison between the proposed BOA methodologies and their grid-based counterparts reveals that BOA based models consistently outperform the Grid-search based models.
  • Dynamic‑Based Proximity Ensemble Method and Partitioned Dynamic‑Based Proximity Ensemble Method exhibit superior performance over state‑of‑the‑art comparative models across cryptocurrency, stock index, and short‑term load forecasting datasets. performance
    Source sentence
    Dynamic-Based Proximity Ensemble Method and Partitioned Dynamic- Based Proximity Ensemble Method are two proposed variations which exhibit superior performance over a state-of-the-art comparative models.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 48 related papers). Coverage is partial, so this is neither replication nor refutation.

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.