Apriori Algorithm-Based Three-Dimensional Mineral Prospectivity Mapping—An Example from Meiling South Area, Xinjiang, China

Q: How were the learning databases of GNB and SVM constructed?

In the study area, the learning databases of GNB and SVM were constructed with a ratio of positive and negative samples at 1:1, totaling 414 individual points. The training and test sets were divided in a ratio of 8:2. All attributes of the spatial voxels were normalized, and the value interval was transformed between 0 and 1 to eliminate the influence of different magnitudes on the accuracy of the model. The parameters and prediction results are summarized in Sections 4.3.1-4.3.3, with Table 6 providing attribute information for the 207 positive samples.

Q: What is the accuracy of positive sample prediction?

The positive sample prediction accuracy reached 0.83, as shown in Figure 8b. This indicates that the GNB model was able to accurately predict positive samples in the test data. The accuracy value suggests that the model performed well in identifying positive samples, which is crucial for applications such as mineralization prediction in the study area. The high accuracy demonstrates the effectiveness of the GNB model in handling the given dataset and its potential for practical use in similar scenarios.

Question

1. What mineral resources are present in the Meiling South study area?

2. What are the main lithological sections of the stratigraphy in the study area?

3. What are the fault structures in the study area?

4. What is the 3D geological model construction method used in the study area?

Accepted Answer

The Meiling South study area in the Hami region of Xinjiang, China, has considerable potential for the development of mineral resources at depth. It includes copper-zinc deposits mainly of volcanogenic massive sulfide (VMS), porphyry, and shallow-forming low-temperature types. The area's geological map shows granite veins, andesitic basalt, andesitic, rhyolite porphyry, and tuff in the southern part. The central area consists mainly of andesitic porphyry and rhyolite porphyry, while the northern part mainly comprises andesitic breccia lava, andesite, and granodiorite. Medium acidic rock bodies are distributed throughout, and significant ore-controlling faults are present in the northwest and northeast directions.

Accepted Answer

The stratigraphy of the study area consists of four main lithological sections from bottom to top. The first section is mainly composed of fractured gray and black-gray basalt and andesite basalt, with small amounts of basaltic andesite, andesite, and volcanic clastic rocks. The second section contains blue-gray and greenish-gray andesite, andesitic tuff, fused tuff, and sunken tuff. The third section contains black-gray andesitic basalt, andesitic basalt, andesitic fused tuff, tuff, and tuff conglomerate. The fourth section contains grayish-white and light red-gray rhyolite, rhyolitic porphyry, conglomerate-bearing rhyolitic sunken tuff, and brecciated tuffaceous lava. The second and fourth lithologic sections and a small number of third lithologic sections are mainly exposed in the study area, while the first lithologic section is not exposed at the surface.

Accepted Answer

The study area has well-developed fault structures, divided into northwest and northeast faults. These faults consist of two sets of conjugate faults, namely north-northeast and north-northwest. The presence of these fault structures indicates the tectonic activity in the region, which has influenced the formation of mineral deposits. The fault structures play a crucial role in controlling the distribution and concentration of minerals, particularly in the volcanic tectonic depression where the copper-zinc polymetallic mine is located. Understanding the fault structures is essential for constructing a conceptual model for the mineralization of VMS-type copper deposits in the Meiling South area. By analyzing the fault structures and their relationship with the ore-controlling elements, researchers can gain insights into the geological processes that have shaped the mineralization patterns in the study area.

Accepted Answer

The 3D geological model construction method used in the study area is the explicit modeling method. This method involves using the SKUA-GOCAD software platform to generate a 3D geological model based on drill hole data and vertical geological cross-sectional diagrams. The data used includes drill hole data containing information on lithology, spatial distribution, and copper content. The lithologies are categorized and combined to form five main lithologies, which are then combined with residual density data and vertical geological cross-sectional diagrams formed by expert recognition. The resulting 3D model represents the ore-controlling elements in the study area, including tuff, pyritic sericitization, andesite, meso-acidic intrusive rock, and lava models, as well as the main fault distribution model.

Accepted Answer

The relationship between Bayes' theorem and GNB model lies in the assumption that the features are independent of one another. In the GNB model, Bayes' theorem is used to calculate the probability of a category label y occurring when a set of features x1 to xn occur. The formula for this relationship is P(y | x1, ..., xn), which represents the probability of y given x1 to xn. Additionally, the GNB model applies a Gaussian Naive Bayesian algorithm to the classification process, assuming that the probability of the features is Gaussian distributed. This method is chosen because Bayesian methods are representative of traditional statistical methods, and it can be used as a broader prediction result in comparative analyses with other prediction methods.

Accepted Answer

The kernel function in SVM plays a crucial role in mapping the original input space to a higher-dimensional feature space. This transformation allows initially linearly indistinguishable samples to become linearly distinguishable in the new kernel space. Common kernels include polynomial, radial basis function (RBF), and sigmoid. The RBF kernel function, for example, is defined as RBF(x1, x2) = exp(-gamma * x1 - x2^2) with gamma being a parameter greater than 0. By utilizing the kernel function, SVM can effectively classify data by locating the decision boundary with the largest margins, leading to improved classification accuracy on unknown datasets.

Accepted Answer

The mean 3D distance of ore-bearing points from pyritic sericitization is 48.52 m. This value represents the average distance between the ore-bearing points and the pyritic sericitization mineralization. The standard deviation is 27.67 m, indicating the variability in the distances. The minimum value is 7.81 m, and the values at 25%, 50%, and 75% of the distribution are 31.49 m, 37.58 m, and 76.23 m, respectively. This information helps in understanding the spatial distribution and proximity of the ore-bearing points to the pyritic sericitization mineralization.

Accepted Answer

The top three attributes correlated with ore-bearing points based on Apriori algorithm are the remaining density of the ore-bearing points > -0.056 g/cm 3 and <=-0.019 g/cm 3, the vertical angle of the point from the lava between 45deg and 90deg, and the 3D distance of the point from the andesite between 0 and 30 m.

Accepted Answer

In the study area, the learning databases of GNB and SVM were constructed with a ratio of positive and negative samples at 1:1, totaling 414 individual points. The training and test sets were divided in a ratio of 8:2. All attributes of the spatial voxels were normalized, and the value interval was transformed between 0 and 1 to eliminate the influence of different magnitudes on the accuracy of the model. The parameters and prediction results are summarized in Sections 4.3.1-4.3.3, with Table 6 providing attribute information for the 207 positive samples.

Accepted Answer

The positive sample prediction accuracy reached 0.83, as shown in Figure 8b. This indicates that the GNB model was able to accurately predict positive samples in the test data. The accuracy value suggests that the model performed well in identifying positive samples, which is crucial for applications such as mineralization prediction in the study area. The high accuracy demonstrates the effectiveness of the GNB model in handling the given dataset and its potential for practical use in similar scenarios.

Accepted Answer

With the increase in training samples, the accuracy of the test set in cross-validation continued to improve and tended to be stable, exceeding 0.9. This indicates that the SVM model's performance on the test set in cross-validation is highly reliable and consistent as the training samples increase. The model's accuracy is a crucial metric for evaluating its effectiveness in predicting outcomes, and in this case, the SVM model demonstrates strong performance in cross-validation scenarios.

Accepted Answer

The comparison of the mineralization prediction results of the ore-bearing voxels determined by Apriori with those determined by the GNB and SVM models revealed some overlap between the mineralization prediction results in terms of spatial location. According to Figures 11 and 7e, the locations of the ore-bearing voxels determined based on Apriori are more closely related to the ore-controlling elements and follow the mineralization rules, while the locations of the ore-bearing voxels determined by SVM and GNB are more in line with the generalization ability of the mathematical model. Despite these differences, the prediction trends of the three methods are generally similar. Among the ore-bearing voxels identified by Apriori, the percentages of voxels predicted to be ore-bearing by SVM and GNB were 21.5% and 34.6% for low-grade ore, and 45.3% and 54.0% for high-grade ore, respectively. The high-grade ore-bearing voxels determined by Apriori align more closely with the mineralization prediction results of the GNB and SVM models, indicating a higher agreement between the methods. Overall, the high and low-grade ore-forming rules identified by Apriori can provide an effective method for further research on the mining of potentially favorable information from multi-source big data.

Accepted Answer

The comparison of prediction results of different mineralization prediction elements in the SVM prediction model shows that the Apriori-identified mineralization prediction elements perform better than the original 19 mineralization prediction elements and the nine mineralization prediction elements provided by Chi-square. The SVM model reaches a steady state with different mineralization prediction elements, and the prediction results through the Apriori-identified mineralization prediction elements have more voxels with prediction probabilities greater than 0.5. However, the study has limitations, such as the limited number of drill holes verified and the representativeness of the SVM model. Further validation and testing on different models are necessary to improve the accuracy and reliability of the mineralization prediction elements.

Accepted Answer

The Apriori algorithm contributes to mineralization prediction by classifying the copper elemental content of ore-bearing points with other attributes. It uses correlation analysis to find effective rules favoring the formation of high-grade and low-grade ore bodies. This method quantitatively evaluates interactions among ore-bearing point attributes, serving as a reference for future identification of mineralization rules in 3D space. Combining the Apriori method with machine learning improves the accuracy of 3D MPM by avoiding excessive input data for SVM training, enhancing evaluation results. The approach can be applied to other regions, aiding in mining potential information and improving 3D MPM accuracy. However, limitations exist, such as the need for comprehensive validation of multiple models and the inability to verify prediction results in more new drill holes due to time and data acquisition factors. Future research will focus on in-depth exploration of mining 3D mineralization information.

Apriori Algorithm-Based Three-Dimensional Mineral Prospectivity Mapping—An Example from Meiling South Area, Xinjiang, China

Chat with Paper

AI Agents for this Paper

Most frequently asked questions

1. What mineral resources are present in the Meiling South study area?

2. What are the main lithological sections of the stratigraphy in the study area?

3. What are the fault structures in the study area?

4. What is the 3D geological model construction method used in the study area?

5. What is the relationship between Bayes' theorem and GNB model?

6. What is the role of kernel function in SVM?

7. What is the mean 3D distance of ore-bearing points from pyritic sericitization?

8. What are the top three attributes correlated with ore-bearing points based on Apriori algorithm?

9. How were the learning databases of GNB and SVM constructed?

10. What is the accuracy of positive sample prediction?

11. What is the accuracy of the SVM model on the test set in cross-validation?

12. How do Apriori, SVM, and GNB models compare in predicting mineralization results?

13. How do different mineralization prediction elements affect the SVM prediction model's accuracy?

14. How does the Apriori algorithm contribute to mineralization prediction?

Citations

Efficient Construction of Voxel Models for Ore Bodies Using an Improved Winding Number Algorithm and CUDA Parallel Computing

References

A Tutorial on Support Vector Machines for Pattern Recognition

Top 10 algorithms in data mining

Chi2: feature selection and discretization of numeric attributes

Constraint-based rule mining in large, dense databases

Support Vector Machines.

Related Papers (5)

An Improved Apriori Algorithm for Association Rules

Improvement of Apriori Algorithm in Association Rule Mining

Improvement of Apriori Algorithm for Association Rules

Typical operation sequence capture based on Apriori algorithm

An improved Apriori Algorithm based on mining association rule