Mapping Mineralization Probabilities using Multilayer Perceptrons

Mineral-potential mapping is the process of combining a set of input maps, each representing a distinct geo-scientific variable, to produce a single map which ranks areas according to their potential to host mineral deposits of a particular type. The maps are combined using a mapping function that must be either provided by an expert (knowledge-driven approach), or induced from sample data (data-driven approach). Current data-driven approaches using multilayer perceptrons (MLPs) to represent the mapping function have several inherent problems: they are highly sensitive to the selection of training data; they do not utilize the contextual information provided by nondeposit data; and there is no objective interpretation of the values output by the MLP. This paper presents a new approach by which MLPs can be trained to output values that can be interpreted strictly as representing posterior probabilities. Other advantages of the approach are that it utilizes all data in the construction of the model, and thus eliminates any dependence on a particular selection of training data. The technique is applied to mapping gold mineralization potential in the Castlemaine region of Victoria, Australia, and results are compared with a method based on estimating probability density functions.     

Probabilistic Prediction Models for Aggregate Quarry Siting

Weights-of-evidence (WofE) and logistic regression techniques were used in a GIS framework to predict the spatial likelihood (prospectivity) of crushed-stone aggregate quarry development. The joint conditional probability models, based on geology, transportation network, and population density variables, were defined using quarry location and time of development data for the New England States, North Carolina, and South Carolina, USA. The Quarry Operation models describe the distribution of active aggregate quarries, independent of the date of opening. The New Quarry models describe the distribution of aggregate quarries when they open. Because of the small number of new quarries developed in the study areas during the last decade, independent New Quarry models have low parameter estimate reliability. The performance of parameter estimates derived for Quarry Operation models, defined by a larger number of active quarries in the study areas, were tested and evaluated to predict the spatial likelihood of new quarry development. Population density conditions at the time of new quarry development were used to modify the population density variable in the Quarry Operation models to apply to new quarry development sites. The Quarry Operation parameters derived for the New England study area, Carolina study area, and the combined New England and Carolina study areas were all similar in magnitude and relative strength. The Quarry Operation model parameters, using the modified population density variables, were found to be a good predictor of new quarry locations. Both the aggregate industry and the land management community can use the model approach to target areas for more detailed site evaluation for quarry location. The models can be revised easily to reflect actual or anticipated changes in transportation and population features.     

Conditional estimates of the number of podiform chromite deposits

A desirable guide for estimating the number of undiscovered mineral deposits is the number of known deposits per unit area from another well-explored permissive terrain. An analysis of the distribution of 805 podiform chromite deposits among ultramafic rocks in 12 subareas of Oregon and 27 counties of California is used to examine and extend this guide. The average number of deposits in this sample of 39 areas is 0.225 deposits per km² of ultramafic rock; the frequency distribution is significantly skewed to the right. Probabilistic estimates can be made by using the observation that the lognormal distribution fits the distribution of deposits per unit area. A further improvement in the estimates is available by using the relationship between the area of ultramafic rock and the number of deposits.The number (N) of exposed podiform chromite deposits can be estimated by the following relationship: log₁₀(N)=–0.194+0.577 log₁₀(area of ultramafic rock). The slope is significantly different from both 0.0 and 1.0. Because the slope is less than 1.0, the ratio of deposits to area of permissive rock is a biased estimator when the area of ultramafic rock is different from the median 93 km². Unbiased estimates of the number of podiform chromite deposits can be made with the regression equation and 80 percent confidence limits presented herein.     

Assessment of Various Fuzzy c-Mean Clustering Validation Indices for Mapping Mineral Prospectivity: Combination of Multifractal Geochemical Model and Mineralization Processes

This paper describes the application of an unsupervised clustering method, fuzzy c-means (FCM), to generate mineral prospectivity models for Cu?±?Au?±?Fe mineralization in the Feizabad District of NE Iran. Various evidence layers relevant to indicators or potential controls on mineralization, including geochemical data, geological–structural maps and remote sensing data, were used. The FCM clustering approach was employed to reduce the dimensions of nine key attribute vectors derived from different exploration criteria. Multifractal inverse distance weighting interpolation coupled with factor analysis was used to generate enhanced multi-element geochemical signatures of areas with Cu?±?Au?±?Fe mineralization. The GIS-based fuzzy membership function MSLarge was used to transform values of the different evidence layers, including geological–structural controls as well as alteration, into a [0–1] range. Four FCM-based validation indices, including Bezdek’s partition coefficient (V_Pc) and partition entropy (V_Pe) indices, the Fukuyama and Sugeno (V_FS) index and the Xie and Beni (V_XB) index, were employed to derive the optimum number of clusters and subsequently generate prospectivity maps. Normalized density indices were applied for quantitative evaluation of the classes of the FCM prospectivity maps. The quantitative evaluation of the results demonstrates that the higher favorability classes derived from V_FS and V_XB (N_d?=?9.19) appear more reliable than those derived from V_Pc and V_Pe (N_d?=?6.12) in detecting existing mineral deposits and defining new zones of potential Cu?±?Au?±?Fe mineralization in the study area.

     

GIS-Based Spatial Analysis of Regional-Scale Structural Controls on Gold Mineralization Along the Bétaré-Oya Shear Zone,Eastern Cameroon

Bétaré-Oya is one of the gold mining districts in the eastern region of Cameroon. Structural controls on gold mineralization were examined along the Bétaré-Oya Shear Zone, providing further clues on favorable areas for mineral exploration. GIS-based methods combining point pattern (i.e., quadrat count, Fry analysis) and distance distribution analysis were employed here to delineate the spatial patterns of known gold deposits and to evaluate their spatial association with geological structures. Results show that the gold deposits in this area are spatially clustered. At the regional scale, the Fry plot indicates an alignment of deposits, suggesting that gold mineralization is controlled by structures oriented NNE–SSW and NE–SW. At the deposit scale, an alignment is also evident, indicating that the mineralization is also controlled by ENE–WSW-trending structures. The cumulative relative frequency distribution of distances from lineament features to gold occurrence points (D_M) and to non-occurrence points (D_N) ratio (D_M/D_N) was used to rank these two major structural trends and their relative importance as mineralization control. The yielded grades show that NE–SW-trending lineaments, akin to P-type structures, play a major role in controlling the gold mineralization in the area compared to other structures. Beyond the goal to foster mineral prospection in the Bétaré-Oya gold district, information yielded in the present study provides relevant criteria for further exploration in the eastern region of Cameroon.

     

Estimation of undiscovered deposits in quantitative mineral resource assessments—examples from Venezuela and Puerto Rico

Quantitative mineral resource assessments used by the United States Geological Survey are based on deposit models. These assessments consist of three parts: (1) selecting appropriate deposit models and delineating on maps areas permissive for each type of deposit; (2) constructing a grade-tonnage model for each deposit model; and (3) estimating the number of undiscovered deposits of each type. In this article, I focus on the estimation of undiscovered deposits using two methods: the deposit density method and the target counting method.In the deposit density method, estimates are made by analogy with well-explored areas that are geologically similar to the study area and that contain a known density of deposits per unit area. The deposit density method is useful for regions where there is little or no data. This method was used to estimate undiscovered low-sulfide gold-quartz vein deposits in Venezuela.Estimates can also be made by counting targets such as mineral occurrences, geophysical or geochemical anomalies, or exploration plays and by assigning to each target a probability that it represents an undiscovered deposit that is a member of the grade-tonnage distribution. This method is useful in areas where detailed geological, geophysical, geochemical, and mineral occurrence data exist. Using this method, porphyry copper-gold deposits were estimated in Puerto Rico.     

A Comparison of the Weights-of-Evidence Method and Probabilistic Neural Networks

The need to integrate large quantities of digital geoscience information to classify locations as mineral deposits or nondeposits has been met by the weights-of-evidence method in many situations. Widespread selection of this method may be more the result of its ease of use and interpretation rather than comparisons with alternative methods. A comparison of the weights-of-evidence method to probabilistic neural networks is performed here with data from Chisel Lake-Andeson Lake, Manitoba, Canada. Each method is designed to estimate the probability of belonging to learned classes where the estimated probabilities are used to classify the unknowns. Using these data, significantly lower classification error rates were observed for the neural network, not only when test and training data were the same (0.02 versus 23%), but also when validation data, not used in any training, were used to test the efficiency of classification (0.7 versus 17%). Despite these data containing too few deposits, these tests of this set of data demonstrate the neural network's ability at making unbiased probability estimates and lower error rates when measured by number of polygons or by the area of land misclassified. For both methods, independent validation tests are required to ensure that estimates are representative of real-world results. Results from the weights-of-evidence method demonstrate a strong bias where most errors are barren areas misclassified as deposits. The weights-of-evidence method is based on Bayes rule, which requires independent variables in order to make unbiased estimates. The chi-square test for independence indicates no significant correlations among the variables in the Chisel Lake–Andeson Lake data. However, the expected number of deposits test clearly demonstrates that these data violate the independence assumption. Other, independent simulations with three variables show that using variables with correlations of 1.0 can double the expected number of deposits as can correlations of –1.0. Studies done in the 1970s on methods that use Bayes rule show that moderate correlations among attributes seriously affect estimates and even small correlations lead to increases in misclassifications. Adverse effects have been observed with small to moderate correlations when only six to eight variables were used. Consistent evidence of upward biased probability estimates from multivariate methods founded on Bayes rule must be of considerable concern to institutions and governmental agencies where unbiased estimates are required. In addition to increasing the misclassification rate, biased probability estimates make classification into deposit and nondeposit classes an arbitrary subjective decision. The probabilistic neural network has no problem dealing with correlated variables—its performance depends strongly on having a thoroughly representative training set. Probabilistic neural networks or logistic regression should receive serious consideration where unbiased estimates are required. The weights-of-evidence method would serve to estimate thresholds between anomalies and background and for exploratory data analysis.     

Calculation of the variance of mean values for blocks in regional resource evaluation studies

Random sampling of a known covariance function can be used during the process of estimating the variance of means or totals for a spatial random variable in blocks of variable size. One advantage of this method is that the precision of any block variance can be determined at the same time as the integral itself. In two-dimensional space this approach yields sufficiently precise results for continuous spatial random variables with exponential, Gaussian, and spherical covariance functions, as well as for point patterns with exponential covariance density or power-law-type, second-order intensity function. Practical examples of application deal with the areal distribution of felsic volcanic rocks and gold deposits in the Abitibi Volcanic Belt, Canadian Shield. The exponential model yields good results in both cases, but, as an overall fit, the fractal (power-law) model performs better in the characterization of the two-dimensional distribution of the gold deposits.     

Saguaro (Carnegiea Gigantea) Densities and Reproduction Over the Northern Sonoran Desert

Casual observations suggest that saguaro populations are densest in southeastern Arizona, although data have not been collected and no study has been done to address this topic. In addition, the topic of reproductive density has similarly never been addressed. Saguaro reproductive output is directly related to the number of adult individuals and the number of branches in the area. Thirty saguaro populations over their U.S. range were sampled to consider two variables: population density and reproductive stem density. Stepwise regression using climate and vegetation (e.g., availability of nurse plants) to predict density yielded tree + Ambrosia cover and maximum July precipitation. Nurse cover, however, is also influenced by summer rain. The partial correlation results suggest that high saguaro densities are linked with high quality nurse cover (i.e., not Larrea tridentata) in addition to summer rainfall. Total cover and mean annual precipitation are the best predictors of reproductive stem density. Mean annual precipitation may be a good predictor of reproductive stem density, because population density is linked with summer rain while branching is linked with winter rain. The plots were also divided into climatic regions. One-way ANOVA shows that the northeast (high winter precipitation) and west (dry) have lower saguaro densities than the southeast (high summer precipitation), while the northeast and southeast both have very high reproductive stem densities relative to the west. The warmer west is less susceptible to periodic freezing mortality, while previous work has shown that the southeast generally regenerates more successfully. Thus in the colder northeast, which is also outside of the primary summer rain and best nurse plant belt, low density populations seem to be maintained only with high reproductive density.     

Application of Radial Basis Functional Link Networks to Exploration for Proterozoic Mineral Deposits in Central Iran

The metallogeny of Central Iran is characterized mainly by the presence of several iron, apatite, and uranium deposits of Proterozoic age. Radial Basis Function Link Networks (RBFLN) were used as a data-driven method for GIS-based predictive mapping of Proterozoic mineralization in this area. To generate the input data for RBFLN, the evidential maps comprising stratigraphic, structural, geophysical, and geochemical data were used. Fifty-eight deposits and 58 ‘nondeposits’ were used to train the network. The operations for the application of neural networks employed in this study involve both multiclass and binary representation of evidential maps. Running RBFLN on different input data showed that an increase in the number of evidential maps and classes leads to a larger classification sum of squared error (SSE). As a whole, an increase in the number of iterations resulted in the improvement of training SSE. The results of applying RBFLN showed that a successful classification depends on the existence of spatially well distributed deposits and nondeposits throughout the study area. An erratum to this article can be found at     

A Modified Weights-of-Evidence Method for Regional Mineral Resource Estimation

Weights-of-evidence (WofE) modeling and weighted logistic regression (WLR) are two methods of regional mineral resource estimation, which are closely related: For example, if all the map layers selected for further analysis are binary and conditionally independent of the mineral occurrences, expected WofE contrast parameters are equal to WLR coefficients except for the constant term that depends on unit area size. Although a good WofE strategy is supposed to achieve approximate conditional independence, a common problem is that the final estimated probabilities are biased. If there are N deposits in a study area and the sum of all estimated probabilities is written as S, then WofE generally results in S > N. The difference S − N can be tested for statistical significance. Although WLR yields S = N, WLR coefficients generally have relatively large variances. Recently, several methods have been developed to obtain WofE weights that either result in S = N, or become approximately unbiased. A method that has not been applied before consists of first performing WofE modeling and following this by WLR applied to the weights. This method results in modified weights with unbiased probabilities satisfying S = N. An additional advantage of this approach is that it automatically copes with missing data on some layers because weights of unit areas with missing data can be set equal to zero as is generally practiced in WofE applications. Some practical examples of application are provided.     

Correlation between assessments of undiscovered oil and gas resources

There is an inbuilt correlation between estimated quantities of oil and gas produced by probabilistic assessments of undiscovered oil and gas resources. Correlation between assessed quantities of oil and gas occurs at every level, whether prospects, plays, basins, continents, or the world. Providing that the oil and gas are assessed in the same run of the computer program, the correlation can be calculated using the paired values of the undiscovered oil and gas volumes calculated in each of the Monte Carlo simulations. It can be seen in the shape and density of a point plot of these paired values. Alternatively, the correlation can be calculated theoretically using an equation written in terms of the data input to the assessment program. These commonly include distributions for the number of accumulations (N), the success rate (s), the accumulation sizes (V), an oil to gas conversion factor, and a proportion of oil to oil plus gas (P _OOG). The cause of the correlation is investigated and explained using point plots and equations for a variety of input distributions. The shape and density of each plot are determined by the pattern of the numbers of oil and gas accumulations, the sizes of the accumulations, and the proportions of oil to oil plus gas. The correlation is caused by the dispersion or spread of the input distributions. It may be positive or negative, tending toward positive as the dispersions ofN, s, andV increase and the dispersion ofP _OOG decreases. The correlation indicates that there is a relationship between the undiscovered oil and gas resources that may be described by fitting a linear regression to a plot of the paired values of the total oil and gas resources. The relationship should be quoted as part of the assessment and might be used to make a better estimate of the value of the undiscovered resources.     

The deposit size frequency method for estimating undiscovered uranium deposits

The deposit size frequency (DSF) method has been developed as a generalization of the method that was used in the National Uranium Resource Evaluation (NURE) program to estimate the uranium endowment of the United States. The DSF method overcomes difficulties encountered during the NURE program when geologists were asked to provide subjective estimates of (1) the endowed fraction of an area judged favorable (factorF) for the occurrence of undiscovered uranium deposits and (2) the tons of endowed rock per unit area (factorT) within the endowed fraction of the favorable area. Because the magnitudes of factorsF andT were unfamiliar to nearly all of the geologists, most geologists responded by estimating the number of undiscovered deposits likely to occur within the favorable area and the average size of these deposits. The DSF method combines factorsF andT into a single factor (F·T) that represents the tons of endowed rock per unit area of the undiscovered deposits within the favorable area. FactorF·T, provided by the geologist, is the estimated number of undiscovered deposits per unit area in each of a number of specified deposit-size classes. The number of deposit-size classes and the size interval of each class are based on the data collected from the deposits in known (control) areas. The DSF method affords greater latitude in making subjective estimates than the NURE method and emphasizes more of the everyday experience of exploration geologists. Using the DSF method, new assessments have been made for the young, organic-rich surficial uranium deposits in Washington and idaho and for the solution-collapse breccia pipe uranium deposits in the Grand Canyon region in Arizona and adjacent Utah.     

Mineral Systems Analysis and Artificial Neural Network Modeling of Chromite Prospectivity in the Western Limb of the Bushveld Complex,South Africa

Radial basis function link neural network (RBFLN) and fuzzy-weights of evidence (fuzzy-WofE) methods were used to assess regional-scale prospectivity for chromite deposits in the Western Limb and the Nietverdiend layered mafic intrusion of the Bushveld Complex in South Africa. Five predictor maps derived from geological, geochemical and geophysical data were processed in a GIS environment and used as spatial proxy for critical processes that were most probably responsible for the formation of the chromite deposits in the study area. The RBFLN was trained using input feature vectors that correspond to known deposits, prospects and non-deposits. The training was initiated by varying the number of radial basis functions (RBFs) and iterations. The results of training the RBFLN provided optimum number of RBFs and iterations that were used for classification of the input feature vectors. The results show that the network classified 73% of the validation deposits into highly prospective areas for chromite deposit, covering 6.5% of the study area. The RBFLN entirely classified all the non-deposit validation points into low prospectivity areas, occupying 86.6% of the study area. In general, the efficiency of the RBFLN in classifying the validation deposits and non-deposits indicates the degree of spatial relationship between the input feature vectors and the training points, which represent chrome mines and prospects. The RBFLN and fuzzy-WofE analyses used in this study are important in guiding identification of regional-scale prospect areas where further chromite exploration can be carried out.     

Asymmetric fuzzy relation analysis method for ranking geoscience variables

A fuzzy relation analysis method is used to derive weights for qualitative variables based on their partial order relations. Two asymmetric measure indexes (incidence coefficient and probability difference) are proposed to measure the asymmetric associations between geoscience variables from which the partial order relations can be constructed. The fuzzy relation analysis method can be implemented in combination with the asymmetric measure indexes leading to new methods for pattern overlay and data integration in mineral potential prediction. Two types of models are proposed and illustrated by two artificial examples: one for predicting targets for undiscovered deposits, and the other for estimating the mineral resource potential of the targets.     

Are Fractal Dimensions of the Spatial Distribution of Mineral Deposits Meaningful?

It has been proposed that the spatial distribution of mineral deposits is bifractal. An implication of this property is that the number of deposits in a permissive area is a function of the shape of the area. This is because the fractal density functions of deposits are dependent on the distance from known deposits. A long thin permissive area with most of the deposits in one end, such as the Alaskan porphyry permissive area, has a major portion of the area far from known deposits and consequently a low density of deposits associated with most of the permissive area. On the other hand, a more equi-dimensioned permissive area, such as the Arizona porphyry permissive area, has a more uniform density of deposits. Another implication of the fractal distribution is that the Poisson assumption typically used for estimating deposit numbers is invalid. Based on datasets of mineral deposits classified by type as inputs, the distributions of many different deposit types are found to have characteristically two fractal dimensions over separate non-overlapping spatial scales in the range of 5–1000 km. In particular, one typically observes a local dimension at spatial scales less than 30–60 km, and a regional dimension at larger spatial scales. The deposit type, geologic setting, and sample size influence the fractal dimensions. The consequence of the geologic setting can be diminished by using deposits classified by type. The crossover point between the two fractal domains is proportional to the median size of the deposit type. A plot of the crossover points for porphyry copper deposits from different geologic domains against median deposit sizes defines linear relationships and identifies regions that are significantly underexplored. Plots of the fractal dimension can also be used to define density functions from which the number of undiscovered deposits can be estimated. This density function is only dependent on the distribution of deposits and is independent of the definition of the permissive area. Density functions for porphyry copper deposits appear to be significantly different for regions in the Andes, Mexico, United States, and western Canada. Consequently, depending on which regional density function is used, quite different estimates of numbers of undiscovered deposits can be obtained. These fractal properties suggest that geologic studies based on mapping at scales of 1:24,000 to 1:100,000 may not recognize processes that are important in the formation of mineral deposits at scales larger than the crossover points at 30–60 km.     

黄河口泥沙淤积估算问题和方法——以钓口河亚三角洲为例

以往在黄河三角洲沉积量的估算中,对沉积物干容重和计算边界等问题不够重视,导致计算结果存在明显出入。本项研究通过广泛收集资料和大量采样分析得到了多种沉积环境下沉积物干容重的计算模型,结合三角洲沉积结构分析和利用地形测量数据,计算了黄河口钓口河流路时期亚三角洲不同时期的沉积量。其中1965年至1974年间钓口河亚三角洲前缘坡脚以内的总淤积量为71.0亿t。其平均干容重为1.36g/cm³。这一干容重用于估算其它亚三角洲沉积量不会造成明显误差。认为忽略三角洲下松软沉积层的压实沉降、三角洲平原相和前缘相中粘性土与非粘性土干容重的差别以及来沙量的测量误差对计算结果影响较小。     

Application of Artificial Neural Network for Gold–Silver Deposits Potential Mapping: A Case Study of Korea

The aim of this study is to analyze hydrothermal gold–silver mineral deposits potential in the Taebaeksan mineralized district, Korea, using an artificial neural network (ANN) and a geographic information system (GIS) environment. A spatial database considering 46 Au and Ag deposits, geophysical, geological, and geochemical data was constructed for the study area using the GIS. The geospatial factors were used with the ANN to analyze mineral potential. The Au and Ag mineral deposits were randomly divided into a training set (70%) to analyze mineral potential using ANN and a test set (30%) to validate predicted potential map. Four different training datasets determined from likelihood ratio and weight of evidence models were applied to analyze and validate the effect of training. Then, the mineral potential index (MPI) was calculated using the trained back-propagation weights, and mineral potential maps (MPMs) were constructed from GIS data for the four training cases. The MPMs were then validated by comparison with the test mineral occurrences. The validation results gave respective accuracies of 73.06, 73.52, 70.11, and 73.10% for the training cases. The comparison results of some training cases showed less sensitive to training data from likelihood ratio than weight of evidence. Overall, the training cases selected from 10% area with low and high index value of MPM_L and MPM_W gave higher accuracy (73.52 and 73.10%) for MPMs than those (73.06 and 70.11%, respectively) from known deposits and 10% area with low index value of MPI_L and MPI_W.     

Artificial Neural Networks for Mineral-Potential Mapping: A Case Study from Aravalli Province,Western India

This paper describes a GIS-based application of a radial basis functional link net (RBFLN) to map the potential of SEDEX-type base metal deposits in a study area in the Aravalli metallogenic province (western India). Available public domain geodata of the study area were processed to generate evidential maps, which subsequently were encoded and combined to derive a set of input feature vectors. A subset of feature vectors with known targets (i.e., either known mineralized or known barren locations) was extracted and divided into (a) a training data set and (b) a validation data set. A series of RBFLNs were trained to determine the network architecture and estimate parameters that mapped the maximum number of validation vectors correctly to their respective targets. The trained RBFLN that gave the best performance for the validation data set was used for processing all feature vectors. The output for each feature vector is a predictive value between 1 and 0, indicating the extent to which a feature vector belongs to either the mineralized or the barren class. These values were mapped to generate a predictive classification map, which was reclassified into a favorability map showing zones with high, moderate and low favorability for SEDEX-type base metal deposits in the study area. The method demarcates successfully high favorability zones, which occupy 6% of the study area and contain 94% of the known base metal deposits.     

Classification of mineral deposits into types using mineralogy with a probabilistic neural network

In order to determine whether it is desirable to quantify mineral-deposit models further, a test of the ability of a probabilistic neural network to classify deposits into types based on mineralogy was conducted. Presence or absence of ore and alteration mineralogy in well-typed deposits were used to train the network. To reduce the number of minerals considered, the analyzed data were restricted to minerals present in at least 20% of at least one deposit type. An advantage of this restriction is that single or rare occurrences of minerals did not dominate the results. Probabilistic neural networks can provide mathematically sound confidence measures based on Bayes theorem and are relatively insensitive to outliers. Founded on Parzen density estimation, they require no assumptions about distributions of random variables used for classification, even handling multimodal distributions. They train quickly and work as well as, or better than, multiple-layer feedforward networks. Tests were performed with a probabilistic neural network employing a Gaussian kernel and separate sigma weights for each class and each variable. The training set was reduced to the presence or absence of 58 reported minerals in eight deposit types. The training set included: 49 Cyprus massive sulfide deposits; 200 kuroko massive sulfide deposits; 59 Comstock epithermal vein gold districts; 17 quartzalunite epithermal gold deposits; 25 Creede epithermal gold deposits; 28 sedimentary-exhalative zinc-lead deposits; 28 Sado epithermal vein gold deposits; and 100 porphyry copper deposits. The most common training problem was the error of classifying about 27% of Cyprus-type deposits in the training set as kuroko. In independent tests with deposits not used in the training set, 88% of 224 kuroko massive sulfide deposits were classed correctly, 92% of 25 porphyry copper deposits, 78% of 9 Comstock epithermal gold-silver districts, and 83% of six quartzalunite epithermal gold deposits were classed correctly. Across all deposit types, 88% of deposits in the validation dataset were correctly classed. Misclassifications were most common if a deposit was characterized by only a few minerals, e.g., pyrite, chalcopyrite,and sphalerite. The success rate jumped to 98% correctly classed deposits when just two rock types were added. Such a high success rate of the probabilistic neural network suggests that not only should this preliminary test be expanded to include other deposit types, but that other deposit features should be added