This study aims to employ supervised Advanced machine learning for the classification of lithological facies from geophysical log data in wells without drilling core samples. For this purpose, a dataset from seven wells in a training set from one of the oil fields in southern Iran has been utilized. This dataset includes natural gamma ray (SGR), corrected gamma ray (CGR), bulk density (RHOB), neutron porosity (NPHI), compressional wave slowness (DTSM), and shear wave slowness (DTCO), which directly influence the classification of geomechanical facies. These parameters are employed as independent variables, while lithological facies serve as the dependent variable for classification. This dataset pertains to depths ranging from 3000 to 4000 meters in the Ilam and Sarvak fractured limestone formations (Bangestan Limestone) of the subsurface. As the title suggests in this article, Initially, through artificial intelligence clustering methods and laboratory studies, these formations were categorized into five distinct lithological facies After this stage, eight supervised machine learning methods were employed, including Regression Logistic, K Neighbors Classifier, Decision Tree Classifier, Random Forest Classifier, Gaussian NB, Gradient Boosting Classifier, Extra Trees Classifier, and Support Vector Machine (SVM), to predict lithological facies in wells without existing classifications. The dataset of these wells underwent training and testing stages with each of these algorithms to construct an appropriate model. As a result, facies labels were predicted. The performance of the models was evaluated using multiple metrics including Accuracy, Precision, F1-Score, and Recall through confusion matrices and ROC curves. The Extra Trees Classifier, Gradient Boosting Classifier, and K Neighbors Classifier showed superior results among these methods. Finally, the model's performance in predicting lithological facies of unseen or out-of-sample wells was presented.

An Intelligent Particle Swarm Classifier (IPSClassifier) is proposed in this paper. This Classifier is described for finding the decision hyperplanes to classify patterns of different classes in the feature space using particle swarm optimization (PSO) algorithm. An intelligent fuzzy controller is designed to improve the performance and efficiency of proposed swarm intelligence based Classifier by adapting three important parameters of PSO (i.e., swarm size, neighborhood size, and constriction coefficient). Three pattern recognition problems with different feature vector dimensions were used to demonstrate the effectiveness of the proposed Classifier. They are the Iris data classification, the Wine data classification, and radar targets classification from backscattered signals. The experimental results show that the performance of the IPS-Classifier is comparable to or better than the k-nearest neighbor (k-NN) and multi-layer perceptron (MLP) Classifiers, which are two conventional Classifiers.

A multi-objective particle swarm optimization (MOPSO) algorithm has been used to design a Classifier which is able to optimize some important pattern recognition indices concurrently. These are Reliability, Score of recognition, and the number of hyperplanes. The proposed Classifier can efficiently approximate the decision hyperplanes for separating the different classes in the feature space and dose not has any over-fitting and over-learning problems.Other swarm intelligence based Classifiers do not have the capability of simultaneous optimizing aforesaid indices and they also may suffer the over-fitting problem.The experimental results show that the proposed multi-objective Classifier can estimate the optimum sets of hyperplanes by approximating the Pareto-front and provide the favorite user's setup for selecting aforesaid indices.

Concepts and laws of physics have been a valuable source of inspiration for engineers to overcome human challenges and problems. Classification is an important example of such problems that play a major role in various fields of engineering sciences. It is shown that discriminative Classifiers tend to outperform their generative counterparts, especially in the presence of sufficient labeled training data. In this paper, we present a new physics-inspired discriminative classification method using minimum potential lines.  To do this, we first consider two groups of fixed point charges (as two classes of data) and a movable Classifier line between them. Then, we find a stable position for the Classifier line by minimizing the total potential integral on the Classifier line due to the two groups of point charges. Surprisingly, it will be shown that the obtained Classifier is actually an uncertainty-based Classifier that minimizes the total uncertainty of the Classifier line. Experimental results show the effectiveness of the proposed approach.

The article suggests an algorithm for regular Classifier ensemble methodology. The proposed methodology is based on possibilistic aggregation to classify samples. The argued method optimizes an objective function that combines environment recognition, multi-criteria aggregation term and a learning term. The optimization aims at learning backgrounds as solid clusters in subspaces of the high-dimensional feature-space via an unsupervised learning including an attribute discrimination component. The unsupervised clustering component assigns degree of typicality to each data pattern in order to identify and reduce the effect of noisy or outlaid data patterns. Then, the suggested technique obtains the best combination parameters for each background. The experimentations on artificial datasets and standard SONAR dataset demonstrate that our Classifier ensemble does better than individual Classifiers in the ensemble.

The concepts of robust classification and intelligently controlling the search process of genetic algorithm (GA) are introduced and integrated with a conventional genetic Classifier for development of a new version of it, which is called Intelligent and Robust GA-Classifier (IRGA-Classifier). It can efficiently approximate the decision hyperplanes in the feature space.It is shown experimentally that the proposed IRGA-Classifier has removed two important weak points of the conventional GA-Classifiers. These problems are the large number of training points and the large number of iterations to achieve a comparable performance with the Bayes Classifier, which is an optimal conventional Classifier.Three examples have been chosen to compare the performance of designed IRGA-Classifier to conventional GA-Classifier and Bayes Classifier. They are the Iris data classification, the Wine data classification, and radar targets classification from backscattered signals. The results show clearly a considerable improvement for the performance of IRGA-Classifier compared with a conventional GA-Classifier.

Pattern recognition systems are widely used in a host of different fields. Due to some reasons such as lack of knowledge about a method based on which the best Classifier is detected for any arbitrary problem, and thanks to significant improvement in accuracy, researchers turn to ensemble methods in almost every task of pattern recognition. Classification as a major task in pattern recognition, have been subject to this transition. The Classifier ensemble which uses a number of base Classifiers is considered as meta-Classifier to learn any classification problem in pattern recognition. Although some researchers think they are better than single Classifiers, they will not be better if some conditions are not met. The most important condition among them is diversity of base Classifiers. Generally in design of multiple Classifier systems, the more diverse the results of the Classifiers, the more appropriate the aggregated result. It has been shown that the necessary diversity for the ensemble can be achieved by manipulation of dataset features, manipulation of data points in dataset, different sub-samplings of dataset, and usage of different classification algorithms. We also propose a new method of creating this diversity. We use Linear Discriminant Analysis to manipulate the data points in dataset. Although the Classifier ensemble produced by proposed method may not always outperform all of its base Classifiers, it always possesses the diversity needed for creation of an ensemble, and consequently it always outperforms all of its base Classifiers on average.