Navigating Data Scarcity in Multivariate Time Series Forecasting: A Hybrid Model Perspective

Data scarcity in Multivariate Time series (MTS) analysis (MTS) poses significant challenges across various fields, including scientific research, business, healthcare, and technology development. Factors such as, high costs of data collection, sensor failures, and the complexity of maintaining synchronized multivariable datasets limits the data availability for MTS analysis. The paper presents an innovative approach for energy consumption forecasting in the context of limited historical data by integrating Graph Neural Networks (GNNs) with Long Short-Term Memory (LSTM) networks. It begins with thorough data preprocessing, enhancing the dataset with temporal features. Sophisticated GNN model is fine-tuned using hyperparameter optimization, bolstered by K-Fold cross-validation for robustness. Synthetic data generated from the trained GNN model enriches the training set. Subsequently, an LSTM network captures temporal dependencies, trained with early stopping techniques to prevent overfitting. Evaluation encompasses metrics like Mean Squared Error (MSE), Mean Absolute Error (MAE), and R2 score. Visual inspection of real versus forecasted data validates the model's performance. By seamlessly integrating data science and machine learning, this approach advances energy consumption forecasting for sustainable energy management.