Comparative study of ten machine learning algorithms for short-term forecasting in gas warning systems.

Area of Science:

• Industrial Engineering
• Data Science
• Machine Learning

Background:

• Limited research exists on selecting practical machine learning (ML) algorithms for real-time industrial short-term forecasting.
• Existing literature lacks comparative analyses of ML algorithm efficiency in specific industrial contexts.

Purpose of the Study:

• To explore and identify more efficient ML algorithms with superior performance for short-term forecasting.
• To address the gap in practical ML algorithm selection for real-time industrial applications.

Main Methods:

• Employs a mixed-methods approach combining literature reviews, a case study, and comparative analysis.
• Evaluates ten widely used ML algorithms on gas warning systems in a case study mine.
• Introduces a novel 2D quadrant diagram for visualizing prediction error and performance assessments.

Main Results:

• Logistic Regression (LR), Random Forest (RF), and Support Vector Machines (SVM) are identified as optimal ML algorithms for short-term forecasting.
• Algorithms are categorized into optimal (LR, RF, SVM), efficient (ARIMA), suboptimal (BP-SOG, KNN, Perceptron), and inefficient (RNN, BP_Resilient, LSTM).
• Case study findings diverge from previous research on the efficiency of ARIMA, KNN, LR, LSTM, and SVM.

Conclusions:

• No single ML algorithm is universally applicable for all short-term forecasting tasks.
• Further investigation into ARIMA, KNN, LR, and LSTM with additional error assessments is recommended.
• The study highlights discrepancies with prior research, necessitating further exploration and raising 20 research questions.