Original Random Forest Research Articles

Improved random forest classification model combined with C5.0 algorithm for vegetation feature analysis in non-agricultural environments

In response to the challenges posed by the high computational complexity and suboptimal classification performance of traditional random forest algorithms when dealing with high-dimensional and noisy non-agricultural vegetation satellite data, this paper proposes an enhanced random forest algorithm based on the C5.0 algorithm. The paper focuses on the Liaohe Plain, selecting two distinct non-agricultural landscape patterns in Shenbei New District and Changtu County as research objects. High-resolution satellite data from GF-2 serves as the experimental dataset. This paper introduces an ensemble feature method based on the bagging concept to improve the original random forest classification model. This method enhances the likelihood of selecting features beneficial to classifying positive class samples, avoiding excessive removal of useful features from negative samples. This approach ensures feature importance and model diversity. The C5.0 algorithm is then employed for feature selection, and the enhanced vegetation index (EVI) is utilized for vegetation coverage estimation. Results indicate that employing a multi-scale parameter selection tool, combined with limited field-measured data, facilitates the identification and classification of plant species in forest landscapes. The C5.0 algorithm effectively selects classification features, minimizing information redundancy. The established object-oriented random forest classification model achieves an impressive accuracy of 94.02% on the aerial imagery for forest classification dataset, with EVI-based vegetation coverage estimation demonstrating high accuracy. In experiments on the same test set, the proposed algorithm attains an average accuracy of 90.20%, outperforming common model algorithms such as bidirectional encoder representation from transformer, FastText, and convolutional neural network, which achieve average accuracies ranging from 84.41 to 88.33% in identifying non-agricultural artificial habitat vegetation features. The proposed algorithm exhibits a competitive edge compared to other algorithms. These research findings contribute scientific evidence for protecting agricultural ecosystems and restoring agricultural ecosystem biodiversity.

Rapid quantitative analysis of petroleum coke properties by laser-induced breakdown spectroscopy combined with random forest based on a variable selection strategy.

Driven by the "double carbon" strategy, petroleum coke short-term demand is growing rapidly as a negative electrode material for artificial graphite. The analysis of petroleum coke physicochemical properties has always been an important part of its research, encompassing significant indicators such as ash content, volatile matter and calorific value. A strategy based on laser-induced breakdown spectroscopy (LIBS) in combination with chemometrics is proposed to realize the rapid and accurate quantification of the above properties. LIBS spectra of 46 petroleum coke samples were collected, and an original random forest (RF) calibration model was constructed by optimizing the pretreatment parameters. The RF calibration model was further optimized based on variable importance measures (VIM) and variable importance in projection (VIP) methods. After variable selection, the elemental spectral lines related to ash content, volatile matter and calorific value modeling were screened out, thus initially exploring the correlation between these properties and elements. Under the optimized spectral pretreatment method, VI threshold and model parameters, the mean relative error (MREP) of the prediction set of ash content, volatile matter and calorific value were 0.0881, 0.0527 and 0.006, the root mean square error (RMSEP) of the prediction set of ash content, volatile matter and calorific value were 0.0471%, 0.6178% and 0.2697 MJ kg-1, respectively, and the determination coefficient (RP2) of the prediction set was 0.9187, 0.9820 and 0.9510, respectively. The combination of LIBS technology and chemometric methods can provide powerful technical means for the analysis and evaluation of the physicochemical properties of petroleum coke.

Prediction of the effluent chemical oxygen demand and volatile fatty acids for anaerobic treatment based on different feature selections machine-learning methods from lab-scale to pilot-scale

The effluent chemical oxygen demand (COD) and volatile fatty acids (VFA) are important indicators for measuring wastewater anaerobic treatment systems. Through lab and pilot experiments on the anaerobic treatment process of high-salt wastewater, combined with different prediction requirements, the feature variables are classified. The effluent-COD and VFA prediction models of five different Machine-learning methods (i.e., back propagation neural network (BPNN), Genetic algorithm optimizes BPNN(GA-BP), support vector machine (SVM), Least squares support vector machine (LSSVM), and random forest (RF)) were established. The modeling results on different dataset (i.e., lab dataset, pilot dataset and mixed dataset) showed that the SVM and LSSVM methods based on statistical learning had the best model prediction performance in lab and pilot datasets, and the RF method performed well in mixed dataset. Furthermore, the genetic algorithm was adopted to optimize the RF model. After optimization of mixed dataset, the effluent-COD prediction (i.e., R2 = 0.966, RMSE/MAE are 61.227/48.584 mg/L) and VFA prediction (i.e., R2 = 0.639, RMSE/MAE are 26.838/21.831 mg/L) have been improved compared with the original RF model (i.e., R2 = 0.918, RMSE/MAE are 86.232/65.03 mg/L of effluent-COD, and R2 = 0.539, RMSE/MAE are 35.489/26.726 mg/L of VFA). This research can provide a reference for the prediction of the anaerobic treatment of actual large-scale industrial wastewater.

Integrating crystal structure and numerical data for predictive models of lithium-ion battery materials: A modified crystal graph convolutional neural networks approach

Machine learning (ML) are currently making significant impact and bringing tremendous opportunities in material science, and has proven to be an effective tool for accelerating the discovery and advancing the development of lithium-ion batteries materials. In this study, we propose a modified version of the crystal convolutional neural network (CGCNN) algorithm, namely mCGCNN. This deep learning model successfully integrates the crystal structure characteristics and the physical/chemical properties of materials, solving the data fusion problem of the current universal models in materials and significantly enhancing the efficiency and accuracy of gravimetric capacity prediction for lithium-ion batteries. In addition, the scale factors α and β are introduced to control the contribution of crystal structure and numerical data of materials to the model, which increases the flexibility and adjustability of the model. A large dataset is extracted from the Materials Project database and analyzed using various ML algorithms, including traditional ML models and deep learning models such as CGCNN and mCGCNN. In comparison to other models, mCGCNN achieves superior performance with MAE of 6.6 mAhg−1 and R2 value of 0.965 in test set. This outperforms the original CGCNN model (MAE: 42.5 mAhg−1, R2: 0.328), XGBoost model (MAE: 23.6 mAhg−1, R2: 0.842), and Random Forest model (MAE: 21.9 mAhg−1, R2: 0.844). Furthermore, the mCGCNN model is also applied to a classification task using the same dataset, achieving an AUC of 0.99 and a total accuracy of 95.5 %. The mCGCNN model not only provides a valuable tool for rapid and accurate screening of high-performance batteries using multi-source material data, but also has significant potential for the discovery of other functional materials, making it widely applicable.

Fast prediction and control of air core in hydrocyclone by machine learning to stabilize operations

Operation stability significantly impacts hydrocyclone separation performance during wastewater treatment, sludge processing, and microplastic removal from water. The air core inside a hydrocyclone is an important indicator of operation stability. This paper presents a machine learning model designed for fast prediction and control of air core profiles. The model is built upon a modified graph neural network (GNN). It is trained by the data generated from a well-established and validated computational fluid dynamics (CFD) model. This GNN-based surrogate model has undergone two modifications to enhance its prediction accuracy. One is data smoothing, to mitigate the adverse effects of the drastic data change in spatial distributions. The other is the loss function modification to incorporate the air core information acquired by the CFD model. The predicted air cores are compared with the original GNN and random forest (RF) against the CFD results. It shows that the new surrogate model can reproduce air profiles and have higher accuracy than other models in predicting spatial distribution results among different error metrics. Furthermore, this surrogate model is combined with the genetic algorithm to optimize the air core. The proposed machine learning model framework offers a promising avenue for the prediction and control of hydrocyclones.

Machine learning for the identification of colour cues to estimate quality parameters of rocket leaves

Computer Vision Systems (CVSs) have proved to be a powerful tool to evaluate the quality of agricultural products in a non-destructive, contactless, sustainable and objective way. Machine learning techniques have proved to simplify the development of CVS and to provide better performance and greater flexibility in matching the requirements of different products and environmental characteristics, but they are often computationally complex and difficult to be understood by humans. It is desirable to develop methods that exploit the benefits of learning and generate simple and fast solutions that are also interpretable by humans. The approach described in this paper analyses a previously developed and effective machine learning model to extract the information useful to develop computationally light and easily understandable algorithms that evaluate the characteristics of interest on rocket leaves. A Random Forest model previously developed to classify visual quality and to estimate chlorophyll and ammonia contents in rocket leaves has been studied to identify a small set of visual characteristics (colours) that correlate with relevant properties of the product. These visual characteristics have been used as input for several simple, fast and easily understandable algorithms that classify visual quality (QL) and estimate chlorophyll and ammonia contents with lower computational complexities compared to the original Random Forest model. Results obtained by these methods are shown and compared with the ones provided by the original Random Forest model. All the algorithms provided a good separation between marketable and non-marketable samples. They required from 1ms to 22 ms to classify a new sample instead of the 25 ms of the original Random Forest model. Additionally, two methods provided good prediction of chlorophyll (R2v = 0.70) and ammonia (R2v = 0.72) contents requiring only 3 ms and 1 ms respectively.

An Improved Random Forest Magnesium Alloy Prediction Method Based on Particle Swarm Optimization

With the development of the material databases’ construction, the use of machine learning methods to process data mining to discover new materials has gradually become a hot topic. The mechanical properties of Mg alloys are related to their components and processing technologies, therefore, it is possible to build prediction model between components, processing technologies and mechanical properties. In order to improve the design efficiency of Mg alloys, using machine learning methods to build a prediction model for the mechanical properties of Mg alloys is of vital importance. To achieve efficient material design, this paper proposed an improved random forest (RF) method based on the Particle Swarm Optimization (PSO) algorithm, and built a Mg alloy performance prediction model. Experiments showed that the accuracy was greatly improved compared with the original RF model, and the prediction accuracy of mechanical properties can reach more than 90%.

Detection of geochemical anomalies related to mineralization using the Random Forest model optimized by the Competitive Mechanism and Beetle Antennae Search

This study aimed to detect the geochemical anomalies related to Fe-mineralization in the Hunjiang area in Jilin Province, China. For this purpose, a Random Forest model was optimized from two aspects of parameters' search and seeking the optimal solution by integrating the “Beetle Antennae Search” and “Competitive Mechanism”. The optimized Random Forest model was compared with the original Random Forest model in the detection of geochemical anomalies related to Fe-mineralization. The results showed that the optimized Random Forest model avoids some limitations of the original Random Forest model, with significant advantages in the detection of mineralization-related anomaly information from exploration geochemical data. It allowed automatic optimization of parameters by integrating two heuristic searching processes. Simultaneously, it focussed more on the quest for the best global optimal solution. Through the case study, it is found that the optimized Random Forest algorithm significantly improved the overall performance for the detection of geochemical anomalies related to Fe-mineralization. The geochemical anomalies detected by the optimized Random Forest model showed close spatial correlation with known iron polymetallic deposits. Therefore, using the optimized Random Forest model to detect geochemical anomalies related to mineralization is of great importance. It also provides a better supporting foundation for geochemical exploration and mineral resource exploration in the Hunjiang area.

Above-Ground Biomass Estimation for Coniferous Forests in Northern China Using Regression Kriging and Landsat 9 Images

Accurate estimation of forest above-ground biomass (AGB) is critical for assessing forest quality and carbon stocks, which can improve understanding of the vegetation growth processes and the global carbon cycle. Landsat 9, the latest launched Landsat satellite, is the successor and continuation of Landsat 8, providing a highly promising data resource for land cover change, forest surveys, and terrestrial ecosystem monitoring. Regression kriging was developed in the study to improve the AGB estimation and mapping using the Landsat 9 image in Wangyedian forest farm, northern China. Multiple linear regression (MLR), support vector machine (SVM), back propagation neural network (BPNN), and random forest (RF) were used as the original models to predict the AGB trends, and the optimal model was used to overlay the results of kriging interpolation based on the residuals to obtain the new AGB predictions. In addition, Landsat 8 images in Wangyedian were used for comparison and verification with Landsat 9. The results showed that all bands of Landsat 8 and Landsat 9 maintained a high degree of uniformity, with positive correlation coefficients ranging from 0.77 to 0.89 (p < 0.01). RF achieved the highest estimation accuracy among all the original models based on the two data sources. However, kriging regression can significantly reduce the estimation error, with the root mean square error (RMSE) decreasing by 55.4% and 51.1%, for Landsat 8 and Landsat 9, respectively, compared to the original RF. Further, the R2 and the lowest RMSE for Landsat 8 were 0.88 and 16.83 t/ha, while, for Landsat 9, they were 0.87 and 17.91 t/ha. The use of regression kriging combined with Landsat 9 imagery has great potential for achieving efficient and highly accurate forest AGB estimates, providing a new reference for long-term monitoring of forest resource dynamics.

Towards convergence rate analysis of random forests for classification

Random forests have been one of the successful ensemble algorithms in machine learning, and the basic idea is to construct a large number of random trees individually and make predictions based on an average of their predictions. The great successes have attracted much attention on theoretical understandings of random forests, mostly focusing on regression problems. This work takes one step towards the convergence rates of random forests for classification. We present the first finite-sample rate O(n−1/(8d+2)) on the convergence of purely random forests for binary classification, which can be improved to be of O(n−1/(3.87d+2)) by considering the midpoint splitting mechanism. We introduce another variant of random forests, which follows Breiman's original random forests but with different mechanisms on splitting dimensions and positions. We present the convergence rate O(n−1/(d+2)(ln⁡n)1/(d+2)) for the variant of random forests, which reaches the minimax rate, except for a factor (ln⁡n)1/(d+2), of the optimal plug-in classifier under the L-Lipschitz assumption. We achieve the tighter convergence rate O(ln⁡n/n) under some assumptions over structural data. This work also takes one step towards the convergence rate of random forests for multi-class learning, and presents the same convergence rates of random forests for multi-class learning as that of binary classification, yet with different constants. We finally provide empirical studies to support the theoretical analysis.

Effects of stopping criterion on the growth of trees in regression random forests.

Random forests are a powerful machine learning tool that capture complex relationships between independent variables and an outcome of interest. Trees built in a random forest are dependent on several hyperparameters, one of the more critical being the node size. The original algorithm of Breiman, controls for node size by limiting the size of the parent node, so that a node cannot be split if it has less than a specified number of observations. We propose that this hyperparameter should instead be defined as the minimum number of observations in each terminal node. The two existing random forest approaches are compared in the regression context based on estimated generalization error, bias-squared, and variance of resulting predictions in a number of simulated datasets. Additionally the two approaches are applied to type 2 diabetes data obtained from the National Health and Nutrition Examination Survey. We have developed a straightforward method for incorporating weights into the random forest analysis of survey data. Our results demonstrate that generalization error under the proposed approach is competitive to that attained from the original random forest approach when data have large random error variability. The R code created from this work is available and includes an illustration.

Random forest-based algorithms for accurate evaluation of ultimate bending capacity of steel tubes

Despite the existence of methods for estimating the behavior of steel circular tubes subjected to pure bending, analytical models are still restricted due to the problem’s complexity and significant nonlinearity. Using the random forest (RF) as the basic model, novel intelligent models are constructed to estimate the ultimate pure bending capacity of circular steel tubes in this study. The RF model’s parameters are optimized using three nature inspired optimization algorithms, namely, the particle swarm optimization (PSO), ant colony optimization (ACO) and whale optimization algorithm (WOA). In the experimental part, a database of 104 tests that comprise 49 and 55 pure bending tests conducted on fabricated and cold-formed steel circular tubes, respectively, are evaluated and utilized to investigate the applicability of the hybrid RF-models. A single RF model is also built for comparative reasons in order to estimate the ultimate pending capacity. Various statistical and graphical measures are used to evaluate the performance of the developed models. The results show that the proposed RF-based nature-inspired algorithms can outperform the original RF predictive model. When the hybrid-RF models were assessed, it was discovered that the RF-WOA performed best. In addition, the influence of each parameter on the prediction findings based on the best RF-model is investigated via sensitivity analysis. Taking into account the overall findings, the hybrid RF-models may be used as powerful tools to predict the ultimate bending capacity of circular steel tubes and may be viable to aid technicians in making proper judgments.

Tunneling-induced settlement prediction using the hybrid feature selection method for feature optimization

Tunneling-induced ground settlement is influenced by a variety of features. In this article, a machine learning model is proposed to predict the ground settlement induced by shield tunneling. A hybrid feature selection method based on the importance degree of variables is first used to select the variables that are most significant to the settlement. Then, from the three perspectives of permutation importance, Sobol’s variance and Shapley additive explanations analysis, the influence of input features to output are quantified and the features are organically combined to construct the subsets for a random forest (RF) model. The monitoring data from a tunnel construction case across the Yellow River is used to evaluate this model. The variable importance measures (VIMs) based RF model with less variables than the original RF model shows a similar performance. Compared to the principal component analysis (PCA) based RF model, VIMs-based RF model shows a better performance while retaining the feature’s physical information, which is critical for future studies that continue to explore the explicit expression of settlement prediction.

Dynamic ensemble selection based improved random forests for fault classification in industrial processes

Fault classification is an important part in industrial process for process monitoring and control. As an ensemble learning approach for classification, random forests has been widely used in different areas. Taking into account the performance of individual decision tree, the diversity between trees and the difference between process data, a k nearest neighbors-hierarchical clustering (KNN-HC) method is proposed in this paper for dynamic ensemble selection (DES) in random forests. In addition, a weighted probability fusion strategy is developed as an alternative of majority voting rule. The experimental evaluation of the proposed method is carried out through the Tennessee Eastman (TE) benchmark process. Results show that the proposed method outperforms three conventional methods, the original random forests (RF) and the static selection based random forests.

Foreign Object Debris Detection for Optical Imaging Sensors Based on Random Forest.

In recent years, aviation security has become an important area of concern as foreign object debris (FOD) on the airport pavement has a huge potential risk to aircraft during takeoff and landing. Therefore, accurate detection of FOD is important to ensure aircraft flight safety. This paper proposes a novel method to detect FOD based on random forest. The complexity of information in airfield pavement images and the variability of FOD make FOD features difficult to design manually. To overcome this challenge, this study designs the pixel visual feature (PVF), in which weight and receptive field are determined through learning to obtain the optimal PVF. Then, the framework of random forest employing the optimal PVF to segment FOD is proposed. The effectiveness of the proposed method is demonstrated on the FOD dataset. The results show that compared with the original random forest and the deep learning method of Deeplabv3+, the proposed method is superior in precision and recall for FOD detection. This work aims to improve the accuracy of FOD detection and provide a reference for researchers interested in FOD detection in aviation.

Improved identification of cotton cultivated areas by applying instance-based transfer learning on the time series of MODIS NDVI

Cotton is an important cash crop and strategic material in the world, as the main source of natural and renewable fiber for textiles. Accurate and timely cotton distribution maps are crucial for monitoring and managing cotton cultivation system. Remotely sensed data have been widely used in cropland mapping, whereas relatively less attention has been paid specifically to cotton mapping partly due to the difficulty in obtaining the training samples over large regions. To resolve this issue, this study introduced the instance-based transfer learning to identify the cotton cultivated areas using the remotely sensed images. The annual time series of normalized difference vegetation index (NDVI) derived from Moderate Resolution Imaging Spectroradiometer (MODIS) images was adopted as input features. The random forest (RF) and long short-term memory (LSTM) algorithms integrated with the Transfer AdaBoost (TrAdaBoost) algorithm were adopted to generate the RF-based and LSTM-based TrAdaBoost approaches, respectively. The experiments were conducted in Arkansas State of the United States, where the cropland data layer (CDL) was available and utilized as a source of auxiliary data. The results showed that both the RF-based and LSTM-based TrAdaBoost performed better than the original RF and LSTM under the condition that the training samples in the target domain were limited. The advantages of transfer learning were much greater when the percentage of training samples from the source domain equals 80%. The two transfer learning approaches were then applied to identify the cotton cultivated areas in Uzbekistan. The cotton areas detected by MODIS images in 2018 were agreed well with the statistics at the sub-national level, with the R2 values of 0.57 and 0.64, respectively. These results demonstrate the potential of the RF-based and LSTM-based TrAdaBoost approaches in generating the cotton distribution maps when there are few samples in the target domain.

Neonatal Adverse Events' Trigger Tool Setup With Random Forest.

This study aimed to develop a trigger tool for detection of neonatal adverse events (AEs) and to validate its effectiveness. Random forest (RF) algorithm was used to build the predictive model by analyzing data from the medical records of 782 neonates in our previous study. Thirteen variables for each patient were used to predict neonatal AEs. Next, the critical variables were selected based on recursive elimination of variables to form the list of triggers. Then, a trigger tool with those triggers was established and tested by reviewing medical records. The positive predictive value of individual triggers and of the entire tool was evaluated. Data from 782 neonates, including 297 patients with and 485 patients without AEs, were collected to build the original RF model. Then, the 6 most important variables, including diarrhea, antibiotic use, fever, death, skin damage, and suspected necrotizing enterocolitis, were selected to establish a neonate-focused trigger tool. The forest with the 6 variables predicted AEs with a sensitivity of 70.7%, a specificity of 92.0%, and an error rate of 16.1%. In a validation study of the trigger tool, 655 neonates with birth weights ≥1500 g were enrolled, and review of their medical records revealed 1709 triggers and 1172 unique AEs. The 3 most common AEs identified were skin damage, iatrogenic diarrhea, and environmental factor-related fever. The total positive predictive value of the trigger tool was 0.686. The neonate-focused trigger tool developed using the RF algorithm efficiently and reliably identifies AEs among hospitalized neonates with birth weights ≥1500 g.

Understanding the Choice of Collection & Delivery Point by the E- Consumer via a Machine Learning Model: Moroccan Case Study

Collection and delivery points (CDP) are evolving rapidly, along with consumption and distribution trends. Understanding the e- consumer's choice of such points is essential for parcel transport companies, as it has a strong impact on their continuity in this competitive environment. the results of the statistical study that we did on our sample show that supermarkets, service stations, and post offices are the most used CDP by Moroccan consumers, especially: full-time employees aged between 18 and 40 and well educated. In addition, this paper focuses on machine learning algorithms such as: Decision Tree, Naive Bayes, K-Nearest Neighbors and Random Forest to create a CDP Classifier. The findings indicate that our model (Optimized Random Forest) outperforms the baseline models in terms of precision, recall, F1 measure and accuracy. In terms of accuracy, our model which achieved 87% outperforms the decision tree model by 9%, the Naive Bayes model by 13%, the KNN model by 6% and the original Random Forest model by 2%. This article has many implications, in terms of theory, it helps researchers to have a clear vision on the application of machine learning in the field of pick-up point delivery, since there are no previous studies applying this technique. On the other hand, it allows parcel transport companies to better understand the behavior of e-consumers using collection and delivery points, before choosing the optimal form of such points. For example, supermarkets are potentially suitable places to set up a CDP, as they are densely located in the city and easily accessible, especially in residential areas.

Identifying peripheral arterial disease in the elderly patientsusing machine-learning algorithms.

Peripheral artery disease (PAD) is a common syndrome in elderly people. Recently, artificial intelligence (AI) algorithms, in particular machine-learning algorithms, have been increasingly used in disease diagnosis. In this study, we designed an effective diagnostic model of PAD in the elderlypatients using artificial intelligence. The study was performed with 539 participants, all over 80years in age, who underwent the measurements of Doppler ultrasonography and ankle-brachial pressure index (ABI). Blood samples were collected. ABI and two machine-learning algorithms (MLAs)-logistic regression and a random forest (RF) model-were established to diagnose PAD. The sensitivity and specificity of the models were analyzed. An additional RF model was designed based on the most significant features of the original RF model and a prospective study was conducted to demonstrate its external validity. Thirteen of the 28 features introduced to the MLAs differed significantly between PAD and non-PAD participants. The respective sensitivities and specificities of logistic regression, RF, and ABI were as follows: logistic regression (81.5%, 83.8%), RF (89.3%, 91.6%) and ABI (85.1%, 84.5%). In the prospective study, the newly designed RF model based on the most significant seven features exhibited an acceptable performance rate for the diagnosis of PAD with 100.0% sensitivity and 90.3% specificity. An RF model was a more effective method than the logistic regression and ABI for the diagnosis of PAD in an elderly cohort.

Prediction of prognosis and survival of patients with gastric cancer by a weighted improved random forest model: an application of machine learning in medicine.

IntroductionIt is essential to predict the survival status of patients based on their prognosis. This can assist physicians in evaluating treatment decisions. Random forest is an excellent machine learning algorithm even without any modification. We propose a new random forest weighting method and apply it to the gastric cancer patient data from the Surveillance, Epidemiology, and End Results (SEER) program. We evaluated the generalization ability of this weighted random forest algorithm on 10 public medical datasets. Furthermore, for the same weighting mode, the difference between using out-of-bag (OOB) data and all training sets as the weighting basis is explored.Material and methods110 697 cases of gastric cancer patients diagnosed between 1975 and 2016 obtained from the SEER database were included in the experiment. In addition, 10 public medical datasets were used for the generalization ability evaluation of this weighted random forest algorithm.ResultsThrough experimental verification, on the SEER gastric cancer patient data, the weighted random forest algorithm improves the accuracy by 0.79% compared with the original random forest. In AUC, macro-averaging increased by 2.32% and micro-averaging increased by 0.51% on average. Among the 10 public datasets, the random forest weighted in accuracy has the best performance on 6 datasets, with an average increase of 1.44% in accuracy and an average increase of 1.2% in AUC.ConclusionsCompared with the original random forest, the weighted random forest model shows a significant improvement in performance, and the effect of using all training data as the weighting basis is better than using OOB data.