Decision Tree Construction Research Articles

A Privacy-Preserving Data Mining Through Comprehensive GNIPP Approach in Sensitive Data Sets

The quick growth of methods for analyzing data and the availability of easily available datasets have made it possible to build a thorough analytics model that can help with support decision-making. In the meantime, protecting personal privacy is crucial. A popular technique for medical evaluation and prediction, decision trees are easy to comprehend and interpret. However, the decision tree construction procedure may reveal personal information about an individual. By keeping the statistical properties intact and limiting the chance of privacy leaking within a reasonable bound, differential privacy offers a formal mathematical definition of privacy. To construct a boosting random forest that preserves privacy, we propose a Gaussian Noise Integrated Privacy Preservation (GNIPP) in this study. To address the issue of personal information breaches, we have designed a unique Gaussian distribution mechanism in GNIPP that enables the nodes with deeper depth to obtain more privacy during the decision tree construction process. We propose a comprehensive boosting technique based on the decision forest's prediction accuracy for assembling multiple decision trees into a forest. Furthermore, we propose an iterative technique to accelerate the assembly of decision trees. After all, we demonstrate through experimentation that the suggested GNIPP outperforms alternative algorithms on two real-world datasets.

Application of SHAP and Multi-Agent Approach for Short-Term Forecast of Power Consumption of Gas Industry Enterprises

Currently, machine learning methods are widely applied in the power industry to solve various tasks, including short-term power consumption forecasting. However, the lack of interpretability of machine learning methods can lead to their incorrect use, potentially resulting in electrical system instability or equipment failures. This article addresses the task of short-term power consumption forecasting, one of the tasks of enhancing the energy efficiency of gas industry enterprises. In order to reduce the risks of making incorrect decisions based on the results of short-term power consumption forecasts made by machine learning methods, the SHapley Additive exPlanations method was proposed. Additionally, the application of a multi-agent approach for the decomposition of production processes using self-generation agents, energy storage agents, and consumption agents was demonstrated. It can enable the safe operation of critical infrastructure, for instance, adjusting the operation modes of self-generation units and energy-storage systems, optimizing the power consumption schedule, and reducing electricity and power costs. A comparative analysis of various algorithms for constructing decision tree ensembles was conducted to forecast power consumption by gas industry enterprises with different numbers of categorical features. The experiments demonstrated that using the developed method and production process factors reduced the MAE from 105.00 kWh (MAPE of 16.81%), obtained through expert forecasting, to 15.52 kWh (3.44%). Examples were provided of how the use of SHapley Additive exPlanation can increase the safety of the electrical system management of gas industry enterprises by improving experts’ confidence in the results of the information system.

HIERARCHICAL NEURAL NETWORKS IN PREDICTING THE PROPERTIES OF OIL AND GAS RESERVOIRS BASED ON WELL AND SEISMIC DATA

This paper describes a technique for hierarchical neural networks based on the nearest neighbor method with preliminary clustering of the original training dataset and construction of a search cluster decision tree. This method is a promising alternative to neural network technologies with deep learning and has quite a few advantages: high learning rate, identification of objects with a low degree of similarity, and the ability to generalize and retrain. As shown by testing the hierarchical neural network method on real data from the West Siberian oil and gas province, predicting the oil saturation in the Vikulov suite interval is much faster and more efficient than inversion approaches to quantitative interpretation of seismic data while achieving fairly similar geological results. This characterizes the proposed method of hierarchical neural networks as an effective tool for the quantitative interpretation of seismic data to solve geological problems.

PyPLIF HIPPOS-aided Construction and Retrospective Validation of Structure-Based Virtual Screening Protocol Targeting VEGFR2

Recently, the discovery of small molecules as inhibitors for vascular endothelial growth factor receptor 2 (VEGFR2) is of timely interest, especially in the area of ocular neovascular diseases. On the other hand, PyPLIF HIPPOS in combination with machine learning techniques has been reported to increase the prediction quality of structure-based virtual screening (SBVS) protocols. The original version of PyPLIF has served in the development of an SBVS protocol that successfully identified novel chalcone derivatives and short peptides as potent inhibitors for acetylcholinesterase. In this short communication, construction and retrospective validation of an SBVS protocol employing PyPLIF HIPPOS targeting VEGFR2 are presented to make it publicly available. The retrospective validation employed 409 active compounds and 24,950 decoys from the enhanced version of the directory of useful decoys. All compounds were docked independently 3 times using AutoDock Vina followed by the identification of the protein-ligand interaction fingerprints (PLIF) employing PyPLIF HIPPOS. The derived ensemble PLIF descriptors were then used in the decision tree construction using a machine-learning technique called recursive partitioning and regression trees. The best decision was then incorporated in the SBVS protocol. The F-measure and enrichment factor values of the SBVS protocol were 0.387 and 76.879, respectively. Hence, the SBVS protocol is readily available to screen small molecules or short peptides.

PseU-KeMRF: A Novel Method for Identifying RNA Pseudouridine Sites.

Pseudouridine is a type of abundant RNA modification that is seen in many different animals and is crucial for a variety of biological functions. Accurately identifying pseudouridine sites within the RNA sequence is vital for the subsequent study of various biological mechanisms of pseudouridine. However, the use of traditional experimental methods faces certain challenges. The development of fast and convenient computational methods is necessary to accurately identify pseudouridine sites from RNA sequence information. To address this, we introduce a novel pseudouridine site prediction model called PseU-KeMRF, which can identify pseudouridine sites in three species, H. sapiens, S. cerevisiae, and M. musculus. Through comprehensive analysis, we selected four RNA coding schemes, including binary feature, position-specific trinucleotide propensity based on single strand (PSTNPss), nucleotide chemical property (NCP) and pseudo k-tuple composition (PseKNC). Then the support vector machine-recursive feature elimination (SVM-RFE) method was used for feature selection and the feature subset was optimized. Finally, the best feature subsets are input into the kernel based on multinomial random forests (KeMRF) classifier for cross-validation and independent testing. As a new classification method, compared with the traditional random forest, KeMRF not only improves the node splitting process of decision tree construction based on multinomial distribution, but also combines the easy to interpret kernel method for prediction, which makes the classification performance better. Our results indicate superior predictive performance of PseU-KeMRF over other existing models, which can prove that PseU-KeMRF is a highly competitive predictive model that can successfully identify pseudouridine sites in RNA sequences.

TEACHING AND LEARNING TO CONSTRUCT DATA-BASED DECISION TREES USING DATA CARDS AS THE FIRST INTRODUCTION TO MACHINE LEARNING IN MIDDLE SCHOOL

This study investigates how 11- to 12-year-old students construct data-based decision trees using data cards for classification purposes. We examine the students' heuristics and reasoning during this process. The research is based on an eight-week teaching unit during which students labeled data, built decision trees, and assessed them using test data. They learned to manually construct decision trees to classify food items as recommendable or not. They utilized data cards with a heuristic that is a simplified form of a machine learning algorithm. We report on evidence that this topic is teachable to middle school students, along with insights for refining our teaching approach and broader implications for teaching machine learning at the school level.

Downscaling soil moisture in regions with high soil heterogeneity: the solution based on ensemble learning with sequential and parallel learner

Soil moisture plays an important role in the water and heat exchanges between the land surface and atmosphere, and it has great importance for agricultural production, ecological planning, and water resources management. Although microwave remote sensing has been widely used in large-scale soil moisture monitoring, the accuracy of the downscaled retrieval results cannot be guaranteed for regions with high vegetation coverage and high soil heterogeneity. To address these challenges, this study built soil moisture indice set based on MODIS and elevation data by calculating the Pearson correlation coefficient (R) and Maximum Information Coefficient (MIC), then constructed decision tree models (Gradient Boosting Decision Tree and Random Forest) about the indice set and low-resolution Soil Moisture Active Passive (SMAP) by using two ensemble learning methods (Bagging and Boosting). The models were applied to the high-resolution soil moisture indices in Jilin Province for the years 2017 to 2020 to generate 1 km-resolution products. In the validation process, Triple Collocation Analysis (TCA), comparison of soil moisture maps with coarse and fine resolution, and in-situ measurements in Lishu County, Tongyu County, and Jilin City were used to evaluate the differences between downscaling soil moisture results and ground observations at network, seasonal and point scales. The results were as follows: (1) The correlation coefficient (R2) calculated by the TCA method was 0.733 (GBDT_36km) > 0.649 (RF_36km), and the error variance was 0.0004 (GBDT_36km) < 0.00058 (RF_36km). (2) R at network scale was 0.798 (GBDT_SM) > 0.662 (RF_SM), RMSE was 0.040 (GBDT_SM) < 0.044 (RF_SM), the point scale R was 0.864 (GBDT_SM) > 0.833 (RF_SM), RMSE was 0.029 (GBDT_SM) < 0.039 (RF_SM). The R in four stages of the growth period was GBDT_SM > RF_SM, RMSE was GBDT_SM < RF_SM. In conclusion, the GBDT and RF models can reliably downscale soil moisture in Jilin Province, and the Boosting ensemble learning method represented by GBDT had a better estimation performance.

Amalgamating an intelligent variant of the gravitational search algorithm with decision trees for email spam detection

In today’s interconnected digital world, email remains a primary mode of digital interaction valued for its suitability in official, academic and business communications. However, despite its utility, email faces significant challenges due to the widespread presence of spam in various forms, such as phishing, suspicious attachments and deceptive content. This issue not only affects the efficiency and security of email communication but also poses a barrier to its reliability. Therefore, it is essential to devise effective methods to tackle the escalating count of spam emails. This research work presents an intriguing methodology to combat the persistent problem of email spam. The proposed method, abbreviated as AIGSADT, is an amalgamation of the intelligent variant of the gravitational search algorithm (IGSA) and decision trees (DTs). The machine learning-based DT algorithm is individually inadequate for dealing with the large amount of constructive data on a certain attribute due to its instability and ineffectiveness. The proposed AIGSADT approach integrates the IGSA algorithm, which is effective in handling large amounts of data to detect email spam. This is achieved by constructing decision trees employing gravitational forces as the means of information transfer through mass agents. Here, the intelligent factor of the IGSA algorithm prevents the trapping of GSA agents in local optima by updating their positions based on information provided by the best and worst agents. The performance of the presented AIGSADT approach is analyzed through experiments conducted on various categories available in the Ling spam dataset. These experimental evaluations aim to analyze the significance of different pre-processing modules across different dataset categories. The comparative analysis indicates the supremacy of the proposed approach compared to state-of-the-art methodologies.

The use of the Isolation Forest model for anomaly detection in measurement data

The subject of the research is the Isolation Forest model, which is a powerful and efficient tool for detecting anomalies in measurement data and outliers, applicable in various fields where ensuring high accuracy and reliability of measurements is important. The goal of the study is to apply the Isolation Forest model to identify unusual or anomalous patterns that differ from typical patterns in the output data. This is achieved by isolating anomalous patterns from normal ones through the construction of multiple different decision trees. The task of the research is to detect outliers in data obtained during the preparation for international comparisons on the state primary standard for mass and volume flow rate of fluid, mass and volume of fluid flowing through a pipeline, by measuring with a сoriolis flowmeter. Data collected during metrological studies undergo processing by the model to detect anomalies. This model analyzes the data and identifies anomalous or outlier values that may indicate systematic or random measurement errors. It enables quick and efficient detection of even the smallest deviations in the data, helping to maintain high accuracy and reliability of measurement results. The main methods for detecting outliers in statistical analysis, which are distribution-independent, are the Grubbs' criterion, interquartile range distribution, and standard deviation. They are sensitive to sample size but are simple and understandable tools. However, the Isolation Forest model also has its limitations, particularly it can be resource-demanding for large datasets. Additionally, it is necessary to consider that using the model requires proper parameter tuning to achieve optimal results. The results of the research include assessment of the Isolation Forest model's effectiveness by comparing it with traditional outlier detection methods. Comparative analysis of the results of different approaches to the same task is an effective method for evaluating the model's performance. Conclusion. The article concludes with the perspective of further research development in this direction. The work will focus on further developing methods for detecting anomalies in measurement data and improving the accuracy and reliability of measurement results in various application fields, which can find broad applications in science and industry.

Novel decision tree models predict the overall survival of patients with submandibular gland cancer.

While the accurate prediction of the overall survival (OS) in patients with submandibular gland cancer (SGC) is paramount for informed therapeutic planning, the development of reliable survival prediction models has been hindered by the rarity of SGC cases. The purpose of this study is to identify key prognostic factors for OS in SGC patients using a large database and construct decision tree models to aid the prediction of survival probabilities in 12, 24, 60 and 120 months. We performed a retrospective cohort study using the Surveillance, Epidemiology and End Result (SEER) program. Demographic and peri-operative predictor variables were identified. The outcome variables overall survival at 12-, 24-, 60, and 120 months. The C5.0 algorithm was utilized to establish the dichotomous decision tree models, with the depth of tree limited within 4 layers. To evaluate the performances of the novel models, the receiver operator characteristic (ROC) curves were generated, and the metrics such as accuracy rate, and area under ROC curve (AUC) were calculated. A total of 1,705, 1,666, 1,543, and 1,413 SGC patients with a follow up of 12, 24, 60 and 120 months and exact survival status were identified from the SEER database. Predictor variables of age, sex, surgery, radiation, chemotherapy, tumor histology, summary stage, metastasis to distant lymph node, and marital status exerted substantial influence on overall survival. Decision tree models were then developed, incorporating these vital prognostic indicators. Favorable consistency was presented between the predicted and actual survival statuses. For the training dataset, the accuracy rates for the 12-, 24-, 60- and 120-month survival models were 0.866, 0.767, 0.737 and 0.797. Correspondingly, the AUC values were 0.841, 0.756, 0.725, and 0.774 for the same time points. Based on the most important predictor variables identified using the large, SEER database, decision tree models were established that predict OS of SGC patients. The models offer a more exhaustive evaluation of mortality risk and may lead to more personalized treatment strategies.

Associated factors leading to misdiagnosis of a combined diagnostic model of different types of strain imaging and conventional ultrasound in evaluation of breast lesions: Selection strategy for using different types of strain imaging in evaluation of breast lesions

ObjectiveTo evaluate the effectiveness of a decision tree that integrates conventional ultrasound (CUS) with two different strain imaging (SI) techniques for diagnosing breast lesions, and to analyze the factors contributing to false negative (FN) and false positive (FP) in the decision tree’s outcomes. Materials and MethodsImaging and clinical data of 796 cases in the training set and 351 cases in the validation set were prospectively collected. A decision tree model that combines two types of SI and CUS was constructed, and its diagnostic performance was analyzed. Univariate analysis and multivariate analysis were applied to identify independent risk factors associated with FP and FN results of the decision tree model. ResultsSize, shape, margin, vascularity, the types of internal calcifications, EI score and VTI pattern were found to be significantly independently associated with the diagnosis of benign and malignant breast lesions. Therefore, size, shape, margin, vascularity, EI score and VTI pattern were used to construct decision tree models. The Tree (EI+VTI) model had the highest AUC. Both in the training and validation groups, the AUC of Tree (EI+VTI) was significantly higher compared with that of EI, VTI, and BI-RADS (all, P < 0.05). Orientation, posterior acoustic features and the types of internal calcifications were significantly positively associated with misdiagnosis results of Tree (EI+VTI) in evaluation of breast lesions (all P < 0.05). ConclusionThe diagnostic model based on a decision tree that integrates two distinct types of SI with CUS enhances the diagnostic accuracy of each method when used individually. This integration lowers the misdiagnosis rate, potentially assisting radiologists in more effective lesion assessments. When applying the decision tree model, attention should be paid to the orientation, posterior acoustic features, and the types of internal calcifications of the lesions.

Value of Real-World Evidence for Treatment Selection: A Case Study in Colon Cancer.

Real-world evidence (RWE)-derived from analysis of real-world data (RWD)-has the potential to guide personalized treatment decisions. However, because of potential confounding, generating valid RWE is challenging. This study demonstrates how to responsibly generate RWE for treatment decisions. We validate our approach by demonstrating that we can uncover an existing adjuvant chemotherapy (ACT) guideline for stage II and III colon cancer (CC)-which came about using both data from randomized controlled trials and expert consensus-solely using RWD. Data from the population-based Netherlands Cancer Registry from a total of 27,056 patients with stage II and III CC who underwent curative surgery were analyzed to estimate the overall survival (OS) benefit of ACT. Focusing on 5-year OS, the benefit of ACT was estimated for each patient using G-computation methods by adjusting for patient and tumor characteristics and estimated propensity score. Subsequently, on the basis of these estimates, an ACT decision tree was constructed. The constructed decision tree corresponds to the current Dutch guideline: patients with stage III or stage II with T stage 4 should receive surgery and ACT, whereas patients with stage II with T stage 3 should only receive surgery. Interestingly, we do not find sufficient RWE to conclude against ACT for stage II with T stage 4 and microsatellite instability-high (MSI-H), a recent addition to the current guideline. RWE, if used carefully, can provide a valuable addition to our construction of evidence on clinical decision making and therefore ultimately affect treatment guidelines. Next to validating the ACT decisions advised in the current Dutch guideline, this paper suggests additional attention should be paid to MSI-H in future iterations of the guideline.

Application of C4.5 Algorithm Using Synthetic Minority Oversampling Technique (SMOTE) and Particle Swarm Optimization (PSO) for Diabetes Prediction

Abstract. Diabetes is the fourth or fifth leading cause of death in most developed countries and an epidemic in many developing countries. Early detection can be a preventive measure that uses a set of existing data to be processed through data mining with a classification process. Purpose: Investigate the efficacy of integrating the C4.5 algorithm with Synthetic Minority Oversampling Technique (SMOTE) and Particle Swarm Optimization (PSO) for improving the accuracy of diabetes prediction models. By employing SMOTE, the study aims to address the class imbalance issue inherent in diabetes datasets, which often contain significantly fewer instances of positive cases (diabetes) than negative cases (non-diabetes). Furthermore, by incorporating PSO, the research seeks to optimize the decision tree construction process within the C4.5 algorithm, enhancing its ability to discern complex patterns and relationships within the data. Methods/Study design/approach: This study proposes the use of the C4.5 classification algorithm by applying the synthetic minority oversampling technique (SMOTE) and particle swarm optimization (PSO) to overcome problems in the diabetes dataset, namely the Pima Indian Diabetes Database (PIDD). Result/Findings: From the research results, the accuracy obtained in applying the C4.5 algorithm without the preprocessing process is 75.97%, while the results of the SMOTE application of the C4.5 algorithm are 80%. Meanwhile, applying the C4.5 algorithm using SMOTE and PSO produces the highest accuracy, with 82.5%. This indicates an increase of 6.53% from the classification results using the C4.5 algorithm. Novelty/Originality/Value: This research contributes novelty by proposing a hybrid approach that combines the C4.5 decision tree algorithm with two advanced techniques, Synthetic Minority Oversampling Technique (SMOTE) and Particle Swarm Optimization (PSO), for the prediction of diabetes. While previous studies have explored the application of machine learning algorithms for diabetes prediction, few have examined the synergistic effects of integrating SMOTE and PSO with the C4.5 algorithm specifically.

Coevolutionary Algorithm for Building Robust Decision Trees under Minimax Regret

In recent years, there has been growing interest in developing robust machine learning (ML) models that can withstand adversarial attacks, including one of the most widely adopted, efficient, and interpretable ML algorithms—decision trees (DTs). This paper proposes a novel coevolutionary algorithm (CoEvoRDT) designed to create robust DTs capable of handling noisy high-dimensional data in adversarial contexts. Motivated by the limitations of traditional DT algorithms, we leverage adaptive coevolution to allow DTs to evolve and learn from interactions with perturbed input data. CoEvoRDT alternately evolves competing populations of DTs and perturbed features, enabling construction of DTs with desired properties. CoEvoRDT is easily adaptable to various target metrics, allowing the use of tailored robustness criteria such as minimax regret. Furthermore, CoEvoRDT has potential to improve the results of other state-of-the-art methods by incorporating their outcomes (DTs they produce) into the initial population and optimize them in the process of coevolution. Inspired by the game theory, CoEvoRDT utilizes mixed Nash equilibrium to enhance convergence. The method is tested on 20 popular datasets and shows superior performance compared to 4 state-of-the-art algorithms. It outperformed all competing methods on 13 datasets with adversarial accuracy metrics, and on all 20 considered datasets with minimax regret. Strong experimental results and flexibility in choosing the error measure make CoEvoRDT a promising approach for constructing robust DTs in real-world applications.

Random subsequence forests

The random forest classifier is widely used in different fields due to its accuracy and robustness. Since its invention, the random forest algorithm is naturally developed for multi-dimensional vectorial data since features can be directly sampled during the decision tree construction procedure. In the context of discrete sequence classification, an explicit feature set is not readily available and we need to employ a feature extraction algorithm before building the random forest. However, such a predefined feature subset may limit the diversity of decision trees since the set of candidate features is composed of all subsequences. As a result, the predictive accuracy of constructed random forest classifier may be reduced. To address this, we propose a new algorithm that is able to directly build a random forest by choosing features from the set of all subsequences adaptively. To improve the running efficiency of our algorithm, the count-suffix tree is utilized to facilitate the fast frequency counting of subsequences so as to accelerate the generation of each randomized decision tree. The experimental results on 15 real datasets show that our method can outperform those state-of-the-art classification algorithms in terms of the predictive accuracy. The source code of our method can be found at: https://github.com/JiaqiWang-dlut/RSForest.

Research on eight machine learning algorithms applicability on different characteristics data sets in medical classification tasks.

With the vigorous development of data mining field, more and more algorithms have been proposed or improved. How to quickly select a data mining algorithm that is suitable for data sets in medical field is a challenge for some medical workers. The purpose of this paper is to study the comparative characteristics of the general medical data set and the general data sets in other fields, and find the applicability rules of the data mining algorithm suitable for the characteristics of the current research data set. The study quantified characteristics of the research data set with 26 indicators, including simple indicators, statistical indicators and information theory indicators. Eight machine learning algorithms with high maturity, low user involvement and strong family representation were selected as the base algorithms. The algorithm performances were evaluated by three aspects: prediction accuracy, running speed and memory consumption. By constructing decision tree and stepwise regression model to learn the above metadata, the algorithm applicability knowledge of medical data set is obtained. Through cross-verification, the accuracy of all the algorithm applicability prediction models is above 75%, which proves the validity and feasibility of the applicability knowledge.

Time analysis of online consumer behavior by decision trees, GUHA association rules, and formal concept analysis

AbstractData analytics plays a significant role within the context of the digital business landscape, particularly concerning online sales, aiming to enhance understanding of customer behaviors in the online realm. We review the recent perspectives and empirical findings from several years of scholarly investigation. Furthermore, we propose combining computational methods to scrutinize online customer behavior. We apply the decision tree construction, GUHA (General Unary Hypotheses Automaton) association rules, and Formal concept analysis for the input dataset of 9123 orders (transactions) of sports nutrition, healthy foods, fitness clothing, and accessories. Data from 2014 to 2021, covering eight years, are employed. We present the empirical discoveries, engage in a critical discourse concerning these findings, and delineate the constraints inherent in the research process. The decision tree for classification of the year’s fourth quarter implies that the most important attributes are country, gross profit category, and delivery. The classification of the morning time implies that the most important attributes are gender and country. Thus, the potential marketing strategies can include heterogeneous conditions for men and women based on these findings. Analyzing the identified groups of customers by concept lattices and GUHA association rules can be valuable for targeted marketing, personalized recommendations, or understanding customer preferences.

An applied study of multi-layer decision tree optimization algorithms in machine learning

Abstract This paper investigates the multi-layer decision tree optimization algorithm in machine learning and explores its efficiency and accuracy in applying complex data classification. The focus is on the optimization strategy of the algorithm when dealing with large datasets with multiple types of records. The research adopts methods such as data set Discretization, fuzzification processing and hierarchical fuzzy decision tree construction algorithm. Through experimental verification, our constructed multilevel fuzzy decision tree performs well in two application scenarios: rainfall prediction and fingerprint identification. In the rainfall prediction experiment, using 5200 sample data, the algorithm achieves a classification accuracy of 61%. In the fingerprint recognition experiment, from 3,000 fingerprint images, a 100% recognition rate was performed, and the rejection rate was 25.2%. The results show that the multi-layer decision tree optimization algorithm can effectively handle large data sets of multiple types and significantly improve the accuracy and efficiency of classification. The algorithm offers high adaptability and accuracy when dealing with continuous type attributes. This study provides a new perspective and methodology for the application of multi-layer decision trees in machine learning, which is of great significance for the future development of data processing and classification techniques.

Two-level optimization by differential evolution in decision tree learning algorithm

Decision tree learning algorithms have a long history, but they are still popular due to their efficiency. Tree construction starts at the root and continues to each leaf node, creating a “near-optimal” tree. One of the key steps in creating a decision tree is the selection of a feature to split at each root node, which affects the classification accuracy. This process can be quite labor intensive. The article proposes a new approach to constructing decision trees based on the use of differential evolution for two-level tree optimization. Differential evolution works in two stages: at the first level, a feature is selected for separation, and at the second level, the threshold value for this feature is optimized. The results of the work were tested on several examples of classification problems.

Lactato sanguíneo e peritoneal na cólica equina: aplicação no atendimento de emergência e construção de uma árvore de decisão

ABSTRACT The objective of this retrospective study was to determine whether blood lactate (BL) and peritoneal lactate (PL) are correlated with heart rate, BL-to-PL ratio and dehydration severity in horses with colic that died within or survived beyond 24 hours of hospital admission, and to construct decision trees aimed at predicting short-term mortality using machine learning algorithms. Medical records (n = 339) from the database of the Veterinary Hospital of the School of Veterinary Medicine and Animal Science, University of São Paulo (FMVZ-USP) of horses admitted with colic were retrieved. Correlations between variables of interest were investigated using the Pearson’s or the Spearman’s correlation coefficient. The level of significance was set at 5% (p&lt;0.05). Algorithms used belong to the WEKA environment and were compared using the paired t test (p&lt;0.05). Peritoneal lactate values were strongly and negatively (p&lt;0.05) correlated with BL:PL ratio, and moderately and positively (p&lt;0.05) correlated with percentage of dehydration. Two decision trees with more than 80% accuracy in predicting mortality were constructed and can be used to inform clinicians about the probability of survival beyond the first 24 hours of hospitalization.