User Purchase Intention Prediction Based on Improved Deep Forest

Widely used deep neural networks currently face limitations in achieving optimal performance for purchase intention prediction due to constraints on data volume and hyperparameter selection.To address this issue,based on the deep forest algorithm and further integrating evolutionary ensemble learning methods,this paper proposes a novel Deep Adaptive Evolutionary Ensemble(DAEE)model.This model introduces model diversity into the cascade layer,allowing it to adaptively adjust its structure to accommodate complex and evolving purchasing behavior patterns.Moreover,this paper optimizes the methods of obtaining feature vectors,enhancement vectors,and prediction results within the deep forest algorithm to enhance the model’s predictive accuracy.Results demonstrate that the improved deep forest model not only possesses higher robustness but also shows an increase of 5.02%in AUC value compared to the baseline model.Furthermore,its training runtime speed is 6 times faster than that of deep models,and compared to other improved models,its accuracy has been enhanced by 0.9%.

面向ICS的CGAN-DEEPFOREST入侵检测

随着工业化与信息化的深度融合,工业控制系统(ICS)的安全问题广受关注,ICS领域出现了许多入侵检测模型.但是,现存模型存在局限性,无法同时解决数据不平衡、分类时间长、小样本检测率低和准确率低的问题.因此,本文提出CGAN-DeepForest入侵检测模型解决上述问题.首先,采用改进的条件生成对抗网络(CGAN)定向扩充数据来改善数据的不平衡性.其次,采用随机森林对平衡后的数据集进行特征提取,降低分类模型训练时间和分类时间.再次,采用深度森林(DeepForest)进行分类,提高小样本检测率和整体准确率,输出分类结果.最后,使用数据集Gas验证模型效果.实验结果表明,本文模型与简单深度森林模型相比准确率整体提升3%,小样本数据NMRI、MFCI、Dos的查全率、查准率、F1分别提高至95%、84%、90%;与随机森林模型相比,准确率整体提高6%,小样本NMRI的查全率提升23%;与深度卷积神经网络相比,准确率接近94%时,模型训练时间和分类时间提高约50%.

Prediction of tree crown width in natural mixed forests using deep learning algorithm

Crown width(CW)is one of the most important tree metrics,but obtaining CW data is laborious and timeconsuming,particularly in natural forests.The Deep Learning(DL)algorithm has been proposed as an alternative to traditional regression,but its performance in predicting CW in natural mixed forests is unclear.The aims of this study were to develop DL models for predicting tree CW of natural spruce-fir-broadleaf mixed forests in northeastern China,to analyse the contribution of tree size,tree species,site quality,stand structure,and competition to tree CW prediction,and to compare DL models with nonlinear mixed effects(NLME)models for their reliability.An amount of total 10,086 individual trees in 192 subplots were employed in this study.The results indicated that all deep neural network(DNN)models were free of overfitting and statistically stable within 10-fold cross-validation,and the best DNN model could explain 69%of the CW variation with no significant heteroskedasticity.In addition to diameter at breast height,stand structure,tree species,and competition showed significant effects on CW.The NLME model(R^(2)=0.63)outperformed the DNN model(R^(2)=0.54)in predicting CW when the six input variables were consistent,but the results were the opposite when the DNN model(R^(2)=0.69)included all 22 input variables.These results demonstrated the great potential of DL in tree CW prediction.

Assessment of China’s forest fi re occurrence with deep learning, geographic information and multisource data

Considerable economic losses and ecological damage can be caused by forest fi res,and compared to suppression,prevention is a much smarter strategy.Accordingly,this study focuses on developing a novel framework to assess forest fi re risks and policy decisions on forest fi re management in China.This framework integrated deep learning algorithms,geographic information,and multisource data.Compared to conventional approaches,our framework featured timesaving,easy implementation,and importantly,the use of deep learning that vividly integrates various factors from the environment and human activities.Information on 96,594 forest fi re points from 2001 to 2019 was collected on Moderate Resolution Imaging Spectroradiometer(MODIS)fi re hotspots from 2001 to 2019 from NASA’s Fire Information Resource Management System.The information was classifi ed into factors such as topography,climate,vegetation,and society.The prediction of forest fi re risk was generated using a fully connected network model,and spatial autocorrelation used to analyze the spatial aggregation correlation of active fi re hotspots in the whole area of China.The results show that high accuracy prediction of fi re risks was achieved(accuracy 87.4%,positive predictive value 87.1%,sensitivity 88.9%,area under curve(AUC)94.1%).Based on this,it was found that Chinese forest fi re risk shows signifi cant autocorrelation and agglomeration both in seasons and regions.For example,forest fi re risk usually raises dramatically in spring and winter,and decreases in autumn and summer.Compared to the national average,Yunnan Province,Guangdong Province,and the Greater Hinggan Mountains region of Heilongjiang Province have higher fi re risks.In contrast,a large region in central China has been recognized as having a long-term,low risk of forest fi res.All forest risks in each region were recorded into the database and could contribute to the forest fi re prevention.The successful assessment of forest fi re risks in this study provides a comprehensive knowledge of fi re risks in China over the last 20 years.Deep learning showed its advantage in integrating multiple factors in predicting forest fi re risks.This technical framework is expected to be a feasible evaluation tool for the occurrence of forest fi res in China.

Dark-Forest:Analysis on the Behavior of Dark Web Traffic via DeepForest and PSO Algorithm

The dark web is a shadow area hidden in the depths of the Internet,which is difficult to access through common search engines.Because of its anonymity,the dark web has gradually become a hotbed for a variety of cyber-crimes.Although some research based on machine learning or deep learning has been shown to be effective in the task of analyzing dark web traffic in recent years,there are still pain points such as low accuracy,insufficient real-time performance,and limited application scenarios.Aiming at the difficulties faced by the existing automated dark web traffic analysis methods,a novel method named Dark-Forest to analyze the behavior of dark web traffic is proposed.In this method,firstly,particle swarm optimization algorithm is used to filter the redundant features of dark web traffic data,which can effectively shorten the training and inference time of the model to meet the realtime requirements of dark web detection task.Then,the selected features of traffic are analyzed and classified using the DeepForest model as a backbone classifier.The comparison experiment with the current mainstream methods shows that Dark-Forest takes into account the advantages of statistical machine learning and deep learning,and achieves an accuracy rate of 87.84%.This method not only outperforms baseline methods such as Random Forest,MLP,CNN,and the original DeepForest in both large-scale and small-scale dataset based learning tasks,but also can detect normal network traffic,tunnel network traffic and anonymous network traffic,which may close the gap between different network traffic analysis tasks.Thus,it has a wider application scenario and higher practical value.

Intelligent Deep Learning Enabled Wild Forest Fire Detection System

The latest advancements in computer vision and deep learning(DL)techniques pave the way to design novel tools for the detection and monitoring of forestfires.In this view,this paper presents an intelligent wild forestfire detec-tion and alarming system using deep learning(IWFFDA-DL)model.The pro-posed IWFFDA-DL technique aims to identify forestfires at earlier stages through integrated sensors.The proposed IWFFDA-DL system includes an Inte-grated sensor system(ISS)combining an array of sensors that acts as the major input source that helps to forecast thefire.Then,the attention based convolution neural network with bidirectional long short term memory(ACNN-BLSTM)model is applied to examine and identify the existence of danger.For hyperpara-meter tuning of the ACNN-BLSTM model,the bacterial foraging optimization(BFO)algorithm is employed and thereby enhances the detection performance.Finally,when thefire is detected,the Global System for Mobiles(GSM)modem transmits messages to the authorities to take required actions.An extensive set of simulations were performed and the results are investigated interms of several aspects.The obtained results highlight the betterment of the IWFFDA-DL techni-que interms of various measures.

Fusion-Based Deep Learning Model for Automated Forest Fire Detection

Earth resource and environmental monitoring are essential areas that can be used to investigate the environmental conditions and natural resources supporting sustainable policy development,regulatory measures,and their implementation elevating the environment.Large-scale forest fire is considered a major harmful hazard that affects climate change and life over the globe.Therefore,the early identification of forest fires using automated tools is essential to avoid the spread of fire to a large extent.Therefore,this paper focuses on the design of automated forest fire detection using a fusion-based deep learning(AFFD-FDL)model for environmental monitoring.The AFFDFDL technique involves the design of an entropy-based fusion model for feature extraction.The combination of the handcrafted features using histogram of gradients(HOG)with deep features using SqueezeNet and Inception v3 models.Besides,an optimal extreme learning machine(ELM)based classifier is used to identify the existence of fire or not.In order to properly tune the parameters of the ELM model,the oppositional glowworm swarm optimization(OGSO)algorithm is employed and thereby improves the forest fire detection performance.A wide range of simulation analyses takes place on a benchmark dataset and the results are inspected under several aspects.The experimental results highlighted the betterment of the AFFD-FDL technique over the recent state of art techniques.

Deep Forest-Based Fall Detection in Internet of Medical Things Environment

This article introduces a new medical internet of things(IoT)framework for intelligent fall detection system of senior people based on our proposed deep forest model.The cascade multi-layer structure of deep forest classifier allows to generate new features at each level with minimal hyperparameters compared to deep neural networks.Moreover,the optimal number of the deep forest layers is automatically estimated based on the early stopping criteria of validation accuracy value at each generated layer.The suggested forest classifier was successfully tested and evaluated using a public SmartFall dataset,which is acquired from three-axis accelerometer in a smartwatch.It includes 92781 training samples and 91025 testing samples with two labeled classes,namely non-fall and fall.Classification results of our deep forest classifier demonstrated a superior performance with the best accuracy score of 98.0%compared to three machine learning models,i.e.,K-nearest neighbors,decision trees and traditional random forest,and two deep learning models,which are dense neural networks and convolutional neural networks.By considering security and privacy aspects in the future work,our proposed medical IoT framework for fall detection of old people is valid for real-time healthcare application deployment.

Deep Learning-Based Classification of Rotten Fruits and Identification of Shelf Life

The freshness of fruits is considered to be one of the essential characteristics for consumers in determining their quality,flavor and nutritional value.The primary need for identifying rotten fruits is to ensure that only fresh and high-quality fruits are sold to consumers.The impact of rotten fruits can foster harmful bacteria,molds and other microorganisms that can cause food poisoning and other illnesses to the consumers.The overall purpose of the study is to classify rotten fruits,which can affect the taste,texture,and appearance of other fresh fruits,thereby reducing their shelf life.The agriculture and food industries are increasingly adopting computer vision technology to detect rotten fruits and forecast their shelf life.Hence,this research work mainly focuses on the Convolutional Neural Network’s(CNN)deep learning model,which helps in the classification of rotten fruits.The proposed methodology involves real-time analysis of a dataset of various types of fruits,including apples,bananas,oranges,papayas and guavas.Similarly,machine learningmodels such as GaussianNaïve Bayes(GNB)and random forest are used to predict the fruit’s shelf life.The results obtained from the various pre-trained models for rotten fruit detection are analysed based on an accuracy score to determine the best model.In comparison to other pre-trained models,the visual geometry group16(VGG16)obtained a higher accuracy score of 95%.Likewise,the random forest model delivers a better accuracy score of 88% when compared with GNB in forecasting the fruit’s shelf life.By developing an accurate classification model,only fresh and safe fruits reach consumers,reducing the risks associated with contaminated produce.Thereby,the proposed approach will have a significant impact on the food industry for efficient fruit distribution and also benefit customers to purchase fresh fruits.

一种改进Deep Forest算法在保险购买预测场景中的应用研究

为了实现保险场景的精准营销,同时充分利用千万级客户和保单历史成交记录的数据特点,本文经热门算法研究和统计理论分析,提出一种基于XGBoost改造的Deep Forest级联算法。该算法采用XGBoost浅层机器学习算法作为Deep Forest级联构建块,同时用AUC-PR标准作为级联构建深度学习不平衡样本评价的自适应过程,并将此算法分别与原有XGBoost算法和原始Deep Forest算法进行性能比较。经实践,上述算法应用投产于保险购买预测场景中,分别比原有XGBoost算法和原Deep Forest算法提高5.5%和2.8%,效果显著;同时提出的浅层学习向基于Deep Forest深度优化操作流程,也为其他类似应用场景提供了实践参考方向。

DeepRanger:覆盖制导的深度森林测试方法

深度学习软件的结构特征与传统软件存在明显差异,因此即使展开了大量测试,依然无法有效衡量测试数据对深度学习软件的覆盖情况和测试充分性,并造成后续使用过程中依然可能存在大量未知错误.深度森林是一种新型深度学习模型,其克服了深度神经网络存在的一些缺点,例如:需要大量训练数据、需要高算力平台、需要大量超参数.但目前还没有相关工作对深度森林的测试方法进行研究.针对深度森林的结构特点,制定了一组由随机森林结点覆盖率RFNC、随机森林叶子覆盖率RFLC、级联森林类型覆盖率CFCC和级联森林输出覆盖率CFOC组成的测试覆盖率评价指标.在此基础上,基于遗传算法设计了覆盖制导的测试数据自动生成方法DeepRanger,可自动生成能有效提高模型覆盖率的测试数据集.为对所提出覆盖指标的有效性进行验证,在深度森林开源项目gcForest和MNIST数据集上设计并进行了一组实验.实验结果表明,所提出的4种覆盖指标均能有效评价测试数据集对深度森林模型的测试充分性.此外,与基于随机选择的遗传算法相比,使用覆盖信息制导的测试数据生成方法DeepRanger能达到更高的模型覆盖率.

WDBM: Weighted Deep Forest Model Based Bearing Fault Diagnosis Method

In the research field of bearing fault diagnosis,classical deep learning models have the problems of too many parameters and high computing cost.In addition,the classical deep learning models are not effective in the scenario of small data.In recent years,deep forest is proposed,which has less hyper parameters and adaptive depth of deep model.In addition,weighted deep forest(WDF)is proposed to further improve deep forest by assigning weights for decisions trees based on the accuracy of each decision tree.In this paper,weighted deep forest model-based bearing fault diagnosis method(WDBM)is proposed.The WDBM is regard as a novel bearing fault diagnosis method,which not only inherits the WDF’s advantages-strong robustness,good generalization,less parameters,faster convergence speed and so on,but also realizes effective diagnosis with high precision and low cost under the condition of small samples.To verify the performance of the WDBM,experiments are carried out on Case Western Reserve University bearing data set(CWRU).Experiments results demonstrate that WDBM can achieve comparative recognition accuracy,with less computational overhead and faster convergence speed.

Research on trend prediction of component stock in fuzzy time series based on deep forest

With the continuous development of machine learning and the increasing complexity of financial data analysis,it is more popular to use models in the field of machine learning to solve the hot and difficult problems in the financial industry.To improve the effectiveness of stock trend prediction and solve the problems in time series data processing,this paper combines the fuzzy affiliation function with stock-related technical indicators to obtain nominal data that can widely reflect the constituent stocks in the case of time series changes by analysing the S&P 500 index.Meanwhile,in order to optimise the current machine learning algorithm in which the setting and adjustment of hyperparameters rely too much on empirical knowledge,this paper combines the deep forest model to train the stock data separately.The experimental results show that(1)the accuracy of the extreme random forest and the accuracy of the multi-grain cascade forest are both higher than that of the gated recurrent unit(GRU)model when the un-fuzzy index-adjusted dataset is used as features for input,(2)the accuracy of the extreme random forest and the accuracy of the multigranular cascade forest are improved by using the fuzzy index-adjusted dataset as features for input,(3)the accuracy of the fuzzy index-adjusted dataset as features for inputting the extreme random forest is improved by 18.89% compared to that of the un-fuzzy index-adjusted dataset as features for inputting the extreme random forest and(4)the average accuracy of the fuzzy index-adjusted dataset as features for inputting multi-grain cascade forest increased by 5.67%.

基于MDFF和DCNN-SVM混合网络的滚动轴承故障诊断研究

针对滚动轴承的故障类型比较多,且具有明显的不确定性,采集的单一的信号往往包含各种冗余信息且容易受到噪声信号的干扰,文章提出基于多域特征融合(multi-domain featurefusion,MDFF)和DCNN-SVM的滚动轴承故障诊断研究。通过对多个传感器采集轴承的振动信号,通过时域、频域和完备自适应噪声集合经验模态分解(complete ensemble empirical mode decomposition with adaptive noise,CEEMDAN)等方法进行特征提取,利用随机森林算法对敏感特征进行筛选,降低特征维度,将优化后的敏感特征值分别输入到DCNN网络中进行自适应特征提取。利用DCNN网络改变各个敏感特征量的权重值,进行综合训练,获得多域融合特征量,输入到支持向量机中进行故障诊断。通过设置多组对比试验可知,提出的方法的识别准确率达到96.82%,比人工-SVM识别准确率提高19.95%,可以有效实现对滚动轴承故障状态的全面诊断,具有一定的应用价值。

Evaluation of deep learning algorithms for landslide susceptibility mapping in an alpine-gorge area:a case study in Jiuzhaigou County

With its high mountains,deep valleys,and complex geological formations,the Jiuzhaigou County has the typical characteristics of a disaster-prone mountainous region in southwestern China.On August 8,2017,a strong Ms 7.0 earthquake occurred in this region,causing some of the mountains in the area to become loose and cracked.Therefore,a survey and evaluation of landslides in this area can help to reveal hazards and take effective measures for subsequent disaster management.However,different evaluation models can yield different spatial distributions of landslide susceptibility,and thus,selecting the appropriate model and performing the optimal combination of parameters is the most effective way to improve susceptibility evaluation.In order to construct an evaluation indicator system suitable for Jiuzhaigou County,we extracted 12 factors affecting the occurrence of landslides,including slope,elevation and slope surface,and made samples.At the core of the transformer model is a self-attentive mechanism that enables any two of the features to be interlinked,after which feature extraction is performed via a forward propagation network(FFN).We exploited its coding structure to transform it into a deep learning model that is more suitable for landslide susceptibility evaluation.The results show that the transformer model has the highest accuracy(86.89%),followed by the random forest and support vector machine models(84.47%and 82.52%,respectively),and the logistic regression model achieves the lowest accuracy(79.61%).Accordingly,this deep learning model provides a new tool to achieve more accurate zonation of landslide susceptibility in Jiuzhaigou County.

Deep learning for predictive mechanical properties of hot-rolled strip in complex manufacturing systems

Higher requirements for the accuracy of relevant models are put throughout the transformation and upgrade of the iron and steel sector to intelligent production.It has been difficult to meet the needs of the field with the usual prediction model of mechanical properties of hotrolled strip.Insufficient data and difficult parameter adjustment limit deep learning models based on multi-layer networks in practical applications;besides,the limited discrete process parameters used make it impossible to effectively depict the actual strip processing process.In order to solve these problems,this research proposed a new sampling approach for mechanical characteristics input data of hot-rolled strip based on the multi-grained cascade forest(gcForest)framework.According to the characteristics of complex process flow and abnormal sensitivity of process path and parameters to product quality in the hot-rolled strip production,a three-dimensional continuous time series process data sampling method based on time-temperature-deformation was designed.The basic information of strip steel(chemical composition and typical process parameters)is fused with the local process information collected by multi-grained scanning,so that the next link’s input has both local and global features.Furthermore,in the multi-grained scanning structure,a sub sampling scheme with a variable window was designed,so that input data with different dimensions can get output characteristics of the same dimension after passing through the multi-grained scanning structure,allowing the cascade forest structure to be trained normally.Finally,actual production data of three steel grades was used to conduct the experimental evaluation.The results revealed that the gcForest-based mechanical property prediction model outperforms the competition in terms of comprehensive performance,ease of parameter adjustment,and ability to sustain high prediction accuracy with fewer samples.

A Survey of the Machine Learning Models for Forest Fire Prediction and Detection

Forest fires are a significant threat to the environment, causing ecological damage, economic losses, and posing a threat to human life. Hence, timely detection and prevention of forest fires are critical to minimizing their impact. In this paper, we review the current state-of-the-art methods in forest fire detection and prevention using predictions based on weather conditions and predictions based on forest fire history. In particular, we discuss different Machine Learning (ML) models that have been used for forest fire detection. Further, we present the challenges faced when implementing the ML-based forest fire detection and prevention systems, such as data availability, model prediction errors and processing speed. Finally, we discuss how recent advances in Deep Learning (DL) can be utilized to improve the performance of current fire detection systems.

Research and Implementation of Cancer Gene Data Classification Based on Deep Learning

Cancer has become a cause of concern in recent years. Cancer genomics is currently a key research direction in the fields of genetic biology and biomedicine. This paper analyzes 5 different types of cancer genes, such as breast, kidney, colon, lung and prostate through machine learning methods, with the goal of building a robust classification model to identify each type of cancer, which will allow us to identify each type of cancer early, thereby reducing mortality.

基于特征换序的SCDF多因子量化选股研究

对深度森林的级联层进行子层再连接,建立了子层连接深度森林(sub-layer connection deep forests,SCDF)的分类算法。级联子层之间通过错误分类信息的传递,使得后续的子层能获得前子层的修正特征,从而有效提升了算法的收敛速度和分类正确率。在深度森林的多粒度扫描部分,利用袋外误差对数据特征的重要性进行排序,使重要性较高的因子可多次参与多粒度扫描,弥补了深度森林多粒度扫描的采样不平衡的缺点,并构建了基于特征换序的SCDF多因子选股模型。实验表明,基于特征换序的SCDF多因子选股模型在2020年1月—2022年1月的沪深300股票的年化收益率为26.47%,累计收益率达到120%,优于深度森林的收益率。