Machine Learning Prediction and Interpretability Analysis of Coal and Gas Outbursts

Abstract

Coal and gas outbursts constitute a major hazard for mining safety, which is critical for the sustainable development of China’s energy industry. Rapid, accurate, and reliable pre-diction is pivotal for preventing and controlling outburst incidents. Nevertheless, the mechanisms driving coal and gas outbursts involve highly complex influencing factors. Four main geological indicators were identified by examining the attributes of these factors and their association to outburst intensity. This study developed a machine learning-based prediction model for outburst risk. Five algorithms were evaluated: K Nearest Neighbors (KNN), Back Propagation (BP), Random Forest (RF), Support Vector Machine (SVM), and eXtreme Gradient Boosting (XGBoost). Model optimization was performed via Bayesian hyperparameter (BO) tuning. Model performance was assessed by the Receiver Operating Characteristic (ROC) curve; the optimized XGBoost model demonstrated strong predictive performance. To enhance model transparency and interpretability, the SHapley Additive exPlanations (SHAP) method was implemented. The SHAP analysis identified geological structure was the most important predictive feature, providing a practical decision support tool for mine executives to prevent and control outburst incidents.

1. Introduction

As the leading global producer and consumer of coal, ensuring mining safety is critical to the sustainable development of China’s energy sector [1,2]. Coal and gas outbursts represent a highly complex gas dynamic hazard in underground coal mining. These phenomena are characterized by the rapid and violent ejection of substantial volumes of coal and gas from the seam within an extremely short period, posing severe threats to mining safety [3,4]. With the initial recorded outburst in France in 1834, more than 40,000 incidents have been recorded in over 20 countries. China has experienced over 13,000 outburst incidents to date, resulting in significant economic damage and casualties [5]. Figure 1 shows statistical data on outbursts in China from 2010 to 2024. Although the number of incidents and fatalities has decreased, the average number of deaths per outburst remains high. In January 2024, a coal and gas outburst at Pingmei No. 12 Mine resulted in 16 fatalities, highlighting the ongoing challenges in preventing and controlling outburst hazards.

The precise forecasting of outbursts is pivotal to against this hazard [6]. With rapid progress in computer technology, artificial neural networks, extreme learning machines, ensemble learning methods, hybrid and evolutionary algorithms, and other artificial intelligence techniques were employed to improve prediction performance [7,8]. For example, Wang [9] developed a chaos mapping and Lévy flight-improved crow search algorithm (ICSA) to optimize a CNN and constructed an outburst prediction model. Liu et al. [10] used a support vector machine (SVM) enhanced by an improved snake optimizer (ISO), in which five refinement strategies were implemented to optimize the original SO algorithm. Nevertheless, most current studies have concentrated on improving model performance, ignoring the interpretability of the results. Many machine learning models operate as ‘black boxes’, exhibiting poor interpretability and making their decision-making logic difficult to comprehend [11]. To improve model explainability, interpretative methods are introduced, aiming to provide technical support and reference for preventing and controlling this hazard. This study is the first systematic application of SHAP for multimodel comparison and global feature interpretation in coal and gas outburst prediction.

2. Materials and Methods

2.1. Analysis of Outburst Influencing Factors

The Pingdingshan mining area is a critical coal production base in China, with 156 outburst incidents recorded by the end of 2024. Figure 2a illustrates the mining area in central Henan Province. Its structural framework is primarily controlled by the Qinling Orogenic Belt, characterized mainly by nearly EW-trending thrust nappes parallel to the Qinling–Dabie tectonic zone [12]. Subjected to multiple phases of tectonic activity [13], the coal-bearing strata have been subjected to significant tectonic compression and shearing, resulting in significant displacement and deformation. The coal structure has been severely damaged, while the high compressive stress has effectively sealed gas within the coal seams. This regional tectonic evolution has provided the following material basis and essential conditions: high tectonic stress, low-strength coal body, and enriched coal seam gas.

2.1.1. Geological Structure

The spatial distribution of outburst incidents exhibits distinct zoning characteristics. Figure 2b shows that the incidents are concentrated within four major outburst zones. Zone I (NW-trending fault-fold outburst zone): Composed of the Niuzhuang syncline, Guozhuang anticline, Niuzhuang thrust fault, and the former No. 11 mine thrust fault (F1 fault). This zone has experienced 16 outbursts. Zone II (Structural intersection zone): Located at the junction of the Xindian fault and Zhangwan fault, with 15 recorded outbursts. Zone III (Compound fold zone): Consisting of the Likou syncline and Guozhuang anticline, this zone has recorded 85 outbursts. Zone IV (upper zone of the Guodishan fault): This zone has recorded 13 outbursts.

2.1.2. In-Situ Stress

These incidents include 15 cases of eruption, 115 cases of extrusion, and 26 cases of outburst. These represent 9.61%, 73.72%, and 16.667% of the total incidents, respectively. As shown in Figure 3, extrusion constitutes the predominant type.

In situ stress represents the superposition of gravitational and tectonic stresses [14]. The measured horizontal principal stress comprises tectonic stress and the horizontal component of gravitational stress, expressed as follows:

$${\sigma }_{h1}={\displaystyle \frac{v}{1-v}}\gamma D$$

(1)

$${\sigma }_{tH}={\sigma }_H-{\displaystyle \frac{v}{1-v}}\gamma D$$

(2)

where σ_h1 is the horizontal component of gravitational stress (MPa); σ_tH is the maximum horizontal tectonic stress (MPa); v denotes the Poisson ratio of the overlying rock mass; γ represents its unit weight (kN/m³); and D corresponds to the burial depth (m).

Drawing on field in situ stress measurements from 50 monitoring points in the Pingdingshan mining area [15,16,17], σ_tH was calculated using Equations (1) and (2). The strata overlying the coal seams are predominantly composed of sandstone and sandy mudstone. Therefore, this study assumes v = 0.25 and γ = 24 kN/m³. As illustrated in

Table 1. σ_tH in different tectonic stress fields (Mpa).

Tectonic Stress Field	Western	Central	Eastern
σtH (Max)	18.54	25.91	56.83
σtH (Min)	5.02	9.86	9.28
σtH (Avg)	10.6	17.72	27.4
Number of accidents	15	16	125

, it is found that σ_tH in the eastern part (including Mines No. 10, No. 12, No. 8, No. 13, and Shoushan No. 1) is significantly higher than that in the western (Mines No. 9, No. 5, No. 7, and No. 11) and central parts (Mines No. 1, No. 2, No. 3, No. 4, and No. 6). Correspondingly, the number of outburst incidents in the eastern part far exceeds those in the western and central parts. This indicates that tectonic stress shows a high correlation with outbursts.

2.1.3. Coal Structure

Tectonic coal is formed through the deformation or destruction of the primary coal structure by prolonged tectonic activity, resulting in a series of characteristics distinct from those of primary structured coal [18,19,20]. Tectonic coal exhibits lower cohesion and strength, making it prone to continuous fragmentation and ejection under in situ stress and gas pressure [21,22,23].

The D, E, F, and G coal seams are the main outburst-prone seams, having experienced 40, 41, 74, and 1 outburst incidents, respectively. Among these, the F coal seam has recorded the highest number of incidents, with outbursts occurring in this seam at Mines No. 4, No. 5, No. 8, No. 9, No. 10, No. 12, No. 13, and Shoushan No. 1. The distribution characteristics of tectonic coal in the F seam are complex, showing significant variations in coal seam damage types and the development thickness of tectonic coal across different areas. Figure 4 and

Table 2. Distribution patterns of tectonic coal in different coal mines.

Coal Mine	Distribution Pattern of Tectonic Coal
Mine No. 11	Tectonic coal is generally not well-developed, with localized occurrences of type III–IV tectonic coal.
Mine No. 9, No. 5	Wrinkle structures are commonly observed, and the thickness of the tectonic coal is stable.
Mine No. 8	The coal seam is severely damaged, and tectonic coal is well-developed.
Mine No. 12, No. 10	Tectonic coal is most pronounced, exhibiting distinct layering, and is locally developed throughout the entire seam.
Western Part of Mine No. 13	Tectonic coal is not well-developed, and type III–IV tectonic coal is developed near faults.
Eastern Part of Mine No. 13 and Shoushan No. 1	The coal seam is relatively severely damaged, and the tectonic coal is relatively thick and exhibits distinct layering.

show that the distribution pattern of tectonic coal correlates well with the characteristics of the eastern mines, which experience a higher frequency and greater intensity of outbursts.

2.1.4. Coal Seam Gas

The maximum gas content data from the D, E, and F coal seams in Mines No. 9, No. 5, No. 6, No. 4, No. 1, No. 2, No. 10, No. 12, No. 8, No. 13, and Shoushan No. 1., along with the corresponding number of outburst incidents, were compiled and are presented in Figure 5 and

Table 3. Number of outbursts in D, E, and F coal seams in different coal mines.

Coal Mines	No. 9	No. 5	No. 6	No. 4	No. 1	No. 10	No. 12	No. 8	Shoushan No. 1	No. 13	Avg Gas Emission per Incident/(m3)
D	-	-	3	11	1	25	-	-	-	-	567.8
E	-	-	-	-	-	17	-	23	1	-	3784.4
F	2	13	-	1	-	8	28	17	1	4	10,869.5

. The gas content in the eastern part of the mining area is generally higher than that in the central and western parts, and this area also experiences the highest number of coal and gas outburst incidents. The maximum gas content in D, E, and F coal seams follows the order of F > E > D seam, which corresponds to the number of outburst incidents and the average outburst gas emission. The distribution patterns of gas content align with the frequency and intensity of outburst incidents.

The occurrence of outbursts is materially based on tectonic coal and gas [24,25,26,27,28,29], whereas in situ stress provides the driving force [30,31]. Geological structures play the dominant controlling role [32,33,34], dictating the characteristics of tectonic coal, the gas reservoir conditions, and the in situ stress field [35,36,37]. This relationship can be expressed as [38]

$$F_{outburst}={\displaystyle \int (G,M,H,\sigma )}$$

(3)

where G, M, H, and σ represents the gas state, coal properties, geological structure, and in situ stress, respectively.

Figure 6 provides a visual representation of this formula. Correspondingly, the risk is heightened by greater gas content, higher in situ stress, increased structural complexity, and reduced coal strength.

2.2. Model Selection

The selection of appropriate machine learning models is critical for accurate prediction. To comprehensively evaluate model performance, a range of classical and contemporary machine learning algorithms were selected for comparative analysis:

(1) K-Nearest Neighbors (KNN): This method is grounded in the principle that similar data points are presumed to share identical class labels [39].

(2) Back Propagation (BP): This architecture employs a multilayer feedforward design, the parameters of which are iteratively optimized via the error backpropagation learning algorithm [40].

(3) Random Forest (RF): This method operates by constructing a multitude of decision trees during training and outputs the class that is the mode of the classes from the individual trees [41].

(4) Support Vector Machine (SVM): A powerful classifier that works by finding the hyperplane in a high-dimensional space that maximizes the margin between different data classes [42].

(5) eXtreme Gradient Boosting (XGBoost): An advanced implementation of gradient boosting ensemble method that sequentially optimizes weak learners to minimize a predefined loss function [43].

2.3. Bayesian Optimization (BO)

Robust tuning methods are essential [44]. BO is an efficient algorithm for hyperparameter tuning [45]. BO employs an intelligent search strategy. It is capable of systematically traversing the hyperparameter space. Consequently, it can rapidly pinpoint hyperparameter configurations that maximize model performance.

The execution of BO primarily involves two key steps:

(1) A Gaussian Process (GP) is employed as a non-parametric surrogate model within the BO framework. It is expressed as

$$f\left(x\right)=gp\left[\mu \left(x\right), k\left(x, x^{\prime }\right)\right]$$

(4)

where μ(x) denotes mean function; k (x, x′) denotes covariance kernel function.

(2) The Expected Improvement (EI) acquisition function governed the selection of successive evaluation points during hyperparameter optimization, navigating the trade-off between venturing into unexplored regions of the parameter space and refining solutions near the current optimum. The formula for the EI function is given by

$$EI(x)=E\left[\max \left(f(x)-f(x_{best}),0\right)\right]$$

(5)

2.4. Interpretability

The SHapley Additive exPlanations (SHAP) method was introduced to interpret the optimal model [46]. This method decomposes and attributes predictive outputs to individual input features via SHAP values, quantifying each feature’s marginal contribution across all instances. The model output g(z) can be expressed as the sum of the base value ϕ₀ (the mean prediction over all samples) and the SHAP value ϕ_i for each feature, expressed as

$$g(z)={\varphi }_0+{\displaystyle \sum _{i=1}^M{\varphi }_i{z}_i}$$

(6)

where M denotes the feature count. z is an indicator function (taking values 0 or 1) denoting the presence of the feature.

The SHAP value is calculated as

$${\varphi }_i={\displaystyle \sum _{S\subseteq \left\{X_1,X_2,\dots X_p\right\}\backslash \left\{X_i\right\}=1}^M\frac{\left|S\right|!\left(p-\left|S\right|-1\right)!}{p!}}\left[f\left(S\cup \left\{X_i\right\}\right)-f\left(S\right)\right]$$

(7)

where p denotes feature count; S represents a subset of features from the complete set {X₁, X₂, …, X_p}; f(S) denotes model output; $f\left(S\cup \left\{X_i\right\}\right)$ denotes the model output after incorporating feature X_i; ${\displaystyle \frac{\left|S\right|!\left(p-\left|S\right|-1\right)!}{p!}}$ represents the weighting factor, which accounts for all possible permutations of the feature subset. For a fixed feature i, the number of possible combinations of the subset S is given by $\left|S\right|!\left(p-\left|S\right|-1\right)!$, considering the total number of permutations p! of all features.

2.5. Evaluation Index

Model performance was assessed via the Receiver Operating Characteristic (ROC) curve and its corresponding Area Under the Curve (AUC) [47]. According to established conventions [48], AUC values are interpreted as follows: a value below 0.5 suggests a model with no discriminative ability; 0.7–0.8 indicates acceptable; 0.8–0.9 indicates excellent; above 0.9 indicates outstanding. The key classification metrics referenced in this analysis are defined by the following formulas:

$$Accuracy={\displaystyle \frac{TP+TN}{TP+TN+FP+FN}}$$

(8)

$$\mathrm{Pr}ecision={\displaystyle \frac{TP}{TP+FP}}$$

(9)

$$Recall={\displaystyle \frac{TP+TN}{TP+FN}}$$

(10)

$$F1 Score=2\times {\displaystyle \frac{\mathrm{Pr}ecision\times Recall}{\mathrm{Pr}ecision+Recall}}$$

(11)

$$FPR={\displaystyle \frac{TP}{TN+FP}}$$

(12)

$$AUC={\displaystyle {\int }_0^1\mathrm{Re}call(FPR)dFPR}$$

(13)

3. Results

3.1. Data Description and Pre-Processing

As shown in

Table 4. Original data of coal and gas outbursts.

Number	X1	X2	X3	X4	X5	X6	X7	X8	X9	X10	X11	Q	Y
1	535	3	3	1	1	5.4	1	0.32	0.66	0.4	11.23	19.7	2
2	522	3	1	3	1	4.8	1	0.35	1.93	0.5	10.17	16	2
3	584	3	1	1	5	3.2	3	0.2	2.38	0.53	12.06	132	3
4	484	3	3	1	3	4.81	1	0.53	5.25	0.5	4.75	12	2
5	566	5	1	1	1	3.5	1	0.51	0.36	0.2	9	30	2
6	463	1	1	3	1	4.81	3	0.26	6.24	0.6	4.8	46	2
7	490	3	1	1	1	4.81	1	0.49	6.24	0.6	7.81	28	2
8	424	1	1	1	1	3.65	3	0.51	0.47	0.27	8.53	6	2
9	535	3	1	1	1	5.2	3	0.29	3.68	0.49	17.14	62	3
10	566	5	3	3	3	3.5	1	0.38	1.04	0.75	8.57	144.6	3
11	563.4	5	1	1	1	3.7	1	0.57	0.76	0.4	8.76	53	3
12	564	5	1	1	1	3.7	1	0.57	0.7	0.38	7.24	0	1
13	485	5	3	5	3	3.3	1	0.11	7.8	1.2	19.65	450	3
14	482	3	1	1	1	3.4	1	0.14	7.6	1.3	18.64	0	1
15	623	3	3	1	3	4.5	1	0.35	0.44	0.3	14.27	22	2
16	584	3	1	1	3	3.2	1	0.24	2.32	0.5	11.86	0	1
17	557.6	1	1	1	1	3.1	3	0.54	0.52	0.4	11.38	43	2
18	557.6	1	3	1	1	3.4	3	0.15	0.78	0.6	18.85	240	3
19	557.6	1	1	1	1	3.2	3	0.46	0.36	0.5	9.48	0	1
20	486	1	1	1	1	3.5	3	0.29	0.33	0.38	12.56	22	2
21	529.8	3	1	3	1	4.3	3	0.67	0.42	0.32	11.48	5	2
22	583	1	1	1	1	4.5	3	0.43	1.15	0.7	9.18	10	2
23	583	1	1	1	1	4.7	1	0.46	1.05	0.66	9.24	0	1
24	533	5	3	3	3	4.1	3	0.23	0.34	0.2	12.57	440	3
25	530	5	1	1	1	4.1	1	0.36	0.32	0.18	12.38	0	1
26	622	3	1	1	1	3	1	0.32	2.31	0.5	20.19	64	3
27	573	1	1	1	1	4.1	3	0.5	0.79	0.34	7.33	16	2
28	537.9	3	1	1	3	5.3	1	0.19	0.22	0.8	23.91	138	3
29	562	3	1	1	1	5.25	1	0.47	4.5	0.5	14.28	12.5	2
30	540	1	1	1	1	4.8	1	0.31	0.62	0.32	12.33	0	1
31	540	1	1	3	1	4.8	3	0.27	0.52	0.3	12.05	8	2
32	457	1	1	1	3	3.5	3	0.15	1.95	0.6	5.09	478	3
33	460	1	1	1	1	3.5	1	0.38	1.38	0.52	4.68	0	1
34	589	3	1	1	1	3.2	3	0.24	2.1	0.6	11.05	4.6	2
35	636.4	5	3	3	5	3.2	3	0.15	3.08	0.46	18.25	396	3
36	584	3	1	1	5	3.2	3	0.25	2.94	0.7	14.18	215	3
37	564.6	5	1	1	1	3.5	1	0.48	0.78	0.6	9.27	44	2
38	480	1	1	1	1	4.81	1	0.53	5.25	0.5	4.75	0	1
39	840	7	3	3	5	4.5	3	0.17	1.15	0.25	23.52	551	3
40	838	5	1	1	1	4.5	1	0.26	1.03	0.25	20.86	0	1
41	566	5	1	3	1	3.5	1	0.51	0.48	0.6	7.93	55	3
42	620	1	1	1	1	3	1	0.34	1.83	0.46	18.75	0	1
43	800	5	1	3	1	3.3	3	0.18	0.42	0.22	15.89	190	3
44	820	5	1	1	1	4.5	1	0.21	1.22	0.28	20.31	0	1
45	614	1	1	1	1	4.5	1	0.55	5.4	0.3	9.87	7	2
46	697	1	1	1	1	4.1	1	0.35	0.58	0.12	15.91	14	2
47	629	3	1	1	1	4.5	1	0.34	0.99	0.18	15.67	32	2
48	490	3	1	1	1	3.2	1	0.51	0.85	0.15	14.32	34	2
49	652	1	1	1	1	4	1	0.54	0.5	0.52	12.03	5	2
50	820	7	1	1	1	4.5	1	0.19	1.22	0.3	24.71	115	3
51	554	1	3	3	1	5.4	1	0.28	0.35	0.3	15.47	27	2
52	482	3	3	1	3	4.81	1	0.53	6.24	0.6	4.75	20	2
53	550	1	1	1	1	5.4	1	0.4	0.28	0.25	10.48	0	1
54	606	1	3	1	1	2	1	0.41	0.34	0.45	7.99	16	2
55	557.6	3	3	1	3	3	3	0.24	1.5	0.5	21.06	180	3
56	563	1	1	1	1	3.5	1	0.55	0.42	0.35	6.21	0	1
57	487	3	3	1	3	4.81	1	0.53	5.25	0.5	4.75	10	2
58	583	1	3	1	1	4.8	1	0.34	1.24	0.75	10.34	20	2
59	580	1	1	1	1	3.4	1	0.26	2.7	0.52	12.27	0	1
60	520	3	1	3	1	4.5	1	0.26	1.38	0.4	12.74	45.5	2

, the original data dataset for coal and gas outbursts consists of 60 sample data points sourced from Reference [49], which originated from Pingdingshan No. 8 Coal Mine, China. The input features comprised multiple influencing factors. To evaluate feature relevance, the Pearson correlation coefficient served as the metric for gauging the predictive linear relationship of individual features to the target. A heatmap of the resulting correlation matrix is presented in Figure 7 for visual analysis.

X₁ belong to in situ stress factors, X₂, X₃, X₄, and X₅ belong to geological structural factors, X₆, X₇, and X₈ belong to coal property factors, and X₉, X₁₀, and X₁₁ belong to coal seam gas factors. The specific indicator definitions are as follows. X₁: coal seam depth (m); X₂: geological structure; X₃: change in coal thickness (m); X₄: soft layer thickness variation; X₅: change in coal seam dip angle; X₆: coal seam thickness; X₇: coal seam soft and fallen; X₈: coal hardiness coefficients; X₉: absolute gas emission volume (m³/min); X₁₀: gas volume fraction (%); X₁₁: initial gas desorption rate (cm³/g); Q: amount of coal discharged in a coal and gas outburst (t); Y: outburst risk level. Risk level Y is categorized as follows: Y = 1 (no risk) when Q = 0; Y = 2 (general risk) when 0 < Q ≤ 50; and Y = 3 (severe risk) when Q > 50.

Considering the nonlinearity of coal and gas outbursts, the mutual information (MI) method is adopted to capture nonlinear relationships among features. The calculation formula is as follows:

$$I(X;Y)={\displaystyle \sum _{x\in X}{\displaystyle \sum _{y\in Y}p(x,y)\log }}({\displaystyle \frac{p(x,y)}{p(x)p(y)}})$$

(14)

As shown in

Table 5. Feature importance scores.

Feature	X1	X2	X3	X4	X5	X6	X7	X8	X9	X10	X11
Pearson	0.14	0.15	0.15	0.14	0.23	0.09	0.15	0.33	0.11	0.2	0.27
MI	0.04	0.39	0.38	0.41	0.49	0.15	0.37	0.41	0.07	0.13	0.29
Score	−0.83	−0.07	−0.09	−0.13	0.89	−1.11	−0.11	1.71	−1.07	−0.08	0.91

, the absolute values of Pearson correlation coefficients and mutual information for all obtained features are normalized via Z-score normalization. A comprehensive score for each feature is then calculated through weighted fusion, using the following formula:

$$Score=\alpha \times Z(\left|pearson\right|)+\beta \times Z(mi)$$

(15)

where α + β = 1. Given the nonlinearity of outbursts, it is assumed that α = 0.3 and β = 0.7.

As shown in Table 5, features exhibiting low correlation with the target variable were removed to simplify the model structure and enhance computational efficiency. Consequently, a total of six features, X₂, X₃, X₅, X₈, X₁₀, and X₁₁, were selected as input variables. Prior to model training, the data were processed using Z-score normalization and scaled to the range [−1, 1] to ensure scale consistency and mitigate the impact of scale differences among features. The expression is as follows:

$$x^{\prime }={\displaystyle \frac{x_i-\overline{x}}{s}}$$

(16)

3.2. Hyperparameter Tuning

The entire dataset was randomly split into training and test sets in a 7:3 ratio. The hyperparameters of the five machine learning algorithms were optimized using BO, with the search ranges listed in

Table 6. Hyperparameters of five algorithms.

Algorithm	Hyperparameter Search Space
KNN	n_neighbors [2, 4, 6,…, 20]
BP	n_hidden [5, 10, 15, 20, 25, 30] learning_rate [0.001, 0.01, 0.1]
RF	max_depth [2, 3, 4,…, 10] n_estimators [50, 100, 150,…, 400] min_samles_leaf [1, 2, 3,…, 10] min_samles_split [2, 4, 6,…, 20]
SVM	C [0.1, 1, 10, 100, 1000] Gamma [0.01, 0.1, 1, 10, 100]
XGBoost	max_depth [1, 2, 3,…, 8] n_estimators [100, 150,…, 500] learning_rate [0.001, 0.01, 0.1, 0.2] reg_alha [0.01, 0.1, 1, 10] reg_lambda [0.01, 1, 10]

.

3.3. Comparative Analysis of Model Accuracy

Accuracy, Precision, Recall, and F1 score are commonly used metrics for classification problems. The higher the value of these indicators, the more preferred the model.

Table 7. Summary of model performance.

Models	AUC	Accuracy	Precision	Recall	F1 Score	AUC (5-CV ± SD)
KNN	0.82	0.87	0.85	0.84	0.85	0.818 ± 0.013
BP	0.84	0.90	0.91	0.88	0.89	0.834 ± 0.015
RF	0.77	0.80	0.82	0.84	0.78	0.765 ± 0.017
SVM	0.85	0.91	0.89	0.89	0.9	0.846 ± 0.012
XGBoost	0.90	0.94	0.93	0.95	0.95	0.892 ± 0.011

summarizes the performance of all models 5-fold stratified cross-validation. As shown in Table 7, XGBoost achieved the highest test AUC (0.90) and Accuracy (0.94), followed closely by SVM.

3.4. Interpretability Analysis

The SHAP method was employed to conduct interpretability analysis on the optimal model. Shapley values were used to assess feature importance of predictive outcomes. Specifically, they quantified the marginal contribution of each feature to the prediction outcome of a given instance. The scatterplot of SHAP values and the feature importance diagram for the XGBoost model are shown in Figure 8. The position on the y-axis determines the importance of the feature, while the Shapley value is located on the x-axis. The color of each instance represents the value of the feature, ranging from low to high.

As can be seen in Figure 8, geological structure is among the most influential features, with wide SHAP distributions. The identification of geological structure by the model as high-impact supports its outburst relevance and interpretability. The analysis results indicate that as the geological structure characteristic value increases, the SHAP value also increases, with the positive contribution rising, indicating a heightened risk level. This aligns with practical observations. Numerous experts and scholars, using methods such as field investigations and empirical analyses, demonstrated that geological structures control the occurrence and distribution of coal and gas outbursts [33,34,50,51,52]. Geological structure occupies a crucial position in determining the occurrence and intensity of coal and gas outbursts.

In addition, features such as coal hardiness coefficients and change in coal thickness also exhibit strong SHAP values, which are important secondary indicators of outburst risk. In contrast, features such as coal seam soft and fallen, change in coal seam dip angle, and soft layer thickness variation were assigned relatively low SHAP values, suggesting a limited impact on model predictions. As shown in Figure 8b, as the characteristic value of coal hardiness coefficients increases, their SHAP value decreases, indicating a greater negative impact and a reduced risk level. All other characteristics have a positive effect on risk level of coal and gas outbursts.

4. Discussion

In recent years, China has experienced frequent coal and gas outburst disasters. Meanwhile, addressing many complex system challenges increasingly relies on the advancement and implementation of intelligent computing methods, including multi-factor predictive modeling. However, current research places excessive emphasis on the models themselves, often improving performance by combining multiple algorithms. This results in overly complex models, making it challenging for coal mine decision-makers to extract key insights from them. In contrast, this study introduces interpretability methods. Beyond comparing the performance of various machine learning models, this study also provides interpretability for model predictions. This approach aids in identifying key factors influencing outbursts, thereby increasing the confidence of mine decision-makers in adopting such models. In accordance with the Coal Mine Safety Regulations, geological exploration and assessment must be strengthened. During both the construction and production phases, if significant changes in geological conditions occur, such as encountering unknown faults or folds, timely supplementary exploration should be conducted. Meanwhile, by monitoring changes in parameters such as gas concentration, emission rate, and mining-induced stress in geologically anomalous zones, advanced warning of outburst risks near geological structural belts can be achieved.

While this study presents a preliminary framework for outburst risk prediction, several limitations remain and suggest directions for future research:

(1) This study is confined to static features and does not incorporate dynamic features, such as microseismic [53], acoustic emission [54], electromagnetic radiation [55], ground penetrating radar [56], and seismic wave [57] computed tomography. Future research should integrate these dynamic features with static ones to enhance the scientific validity and predictive accuracy of the framework.

(2) Future work could employ more advanced feature engineering techniques, including graph-based feature interaction and automated feature selection, to better capture nonlinear relationships among predictors and further enhance model performance.

(3) Beyond the models evaluated here, other advanced architectures, including deep learning models such as Temporal Convolutional network (TCN) and Generative Adversarial network (GAN), could be investigated to offer new methodological insights for out-burst risk prediction.

5. Conclusions

The prevention of coal and gas outbursts is critical for mining safety and, by extension, energy sustainability. Thus, enhancing the accuracy of outburst prediction is of paramount importance. By modeling the association between outburst drivers and outburst intensity, a data-informed predictive framework was constructed and rigorously evaluated, establishing a machine learning approach for outburst risk prediction.

(1) Incorporating an excessive number of indicators can degrade model performance. Analyzing and visualizing the correlations between influencing factors and the target variable via Pearson’s correlation coefficient and mutual information enables the identification of key predictors.

(2) A comparative evaluation framework was implemented, in which all models were optimized via BO and assessed using ROC curve analysis. Among them, XGBoost delivered the most robust predictive performance.

(3) Analysis of SHAP values revealed geological structure occupies a crucial position in coal and gas outburst relevance and interpretability. Coal hardiness coefficients and change in coal thickness also exhibit strong SHAP values, which are important secondary indicators of outburst risk.