Forecasting Commodity Prices using Machine Learning

Abstract

This research endeavor focuses on forecasting price movements across 14 diverse commodities by leveraging technical analysis and comprehensive dataset correlations. The dataset integrates pivotal economic indicators and historical commodity prices, covering a wide spectrum from natural gas to agricultural products. Key technical indicators, such as lagged values, moving averages, MACD, historical volatility, and standard deviation, are used to predict percentage changes in commodity prices over time. The methodology underscores meticulous preprocessing steps, crafting technical features to capture intricate patterns and interrelationships within the dataset. These features are derived from a blend of time series analysis, statistical exploration, and machine learning techniques. An extensive suite of machine learning models is employed, including SVM, Decision Trees, KNN, XGBoost, Random Forests, Linear Regression, Gradient Boosting, VAR, VARIMA, VARMA, GRU, LSTM, and Extra Trees. Thorough evaluation with diverse metrics provides a comprehensive understanding of their efficacy in price prediction. Further enhancements include Isolation Forest for outlier detection, feature standardization with Standards Caler, and hierarchical clustering techniques, optimizing dataset analysis and model performance. Ultimately, this research aims to develop a precise predictive model merging technical analysis with economic indicators for commodity price forecasting. The potential insights could significantly impact decision-making in commodity trading and financial markets, presenting valuable contributions to this evolving landscape.