Balanced Training Set Research Articles

Probabilistic Contrastive Learning for Long-Tailed Visual Recognition.

Long-tailed distributions frequently emerge in real-world data, where a large number of minority categories contain a limited number of samples. Such imbalance issue considerably impairs the performance of standard supervised learning algorithms, which are mainly designed for balanced training sets. Recent investigations have revealed that supervised contrastive learning exhibits promising potential in alleviating the data imbalance. However, the performance of supervised contrastive learning is plagued by an inherent challenge: it necessitates sufficiently large batches of training data to construct contrastive pairs that cover all categories, yet this requirement is difficult to meet in the context of class-imbalanced data. To overcome this obstacle, we propose a novel probabilistic contrastive (ProCo) learning algorithm that estimates the data distribution of the samples from each class in the feature space, and samples contrastive pairs accordingly. In fact, estimating the distributions of all classes using features in a small batch, particularly for imbalanced data, is not feasible. Our key idea is to introduce a reasonable and simple assumption that the normalized features in contrastive learning follow a mixture of von Mises-Fisher (vMF) distributions on unit space, which brings two-fold benefits. First, the distribution parameters can be estimated using only the first sample moment, which can be efficiently computed in an online manner across different batches. Second, based on the estimated distribution, the vMF distribution allows us to sample an infinite number of contrastive pairs and derive a closed form of the expected contrastive loss for efficient optimization. Other than long-tailed problems, ProCo can be directly applied to semi-supervised learning by generating pseudo-labels for unlabeled data, which can subsequently be utilized to estimate the distribution of the samples inversely. Theoretically, we analyze the error bound of ProCo. Empirically, extensive experimental results on supervised/semi-supervised visual recognition and object detection tasks demonstrate that ProCo consistently outperforms existing methods across various datasets.

Imbalanced Diagnosis Scheme for Incipient Rotor Faults in Inverter-Fed Induction Motors

Recently, fault diagnosing supervised classifiers have been widely proposed to diagnose both electric and mechanical faults in induction motors (IM). However, many of them require a large amount of data, which implies a great effort required for processing fault-related features and building the training set. Furthermore, in real-world datasets, it is required to deal with highly skewed data distributions, also known as class imbalance, which is a limiting issue and can misguide the tuning of machine learning algorithms. Resampling techniques based on a synthetic generation of minority class observations aim to address this problem. Last but not least is the fact that inverter-fed IM introduces undesired harmonics in the monitoring signal altering the diagnosis patterns. This diagnosis scheme is evaluated on experimental imbalanced data oriented to deal with the diagnosis of a rotor in situations where it is fed with an inverter. The results show how this imbalanced approach determines the actual diagnosis performance on a small amount of data. The experimental results demonstrate that balanced training sets built with class balancing techniques improve the classifier and therefore its performance for diagnosing incipient rotor faults in inverter-fed IM with studied and interpretable features recently proposed in this field of study.

Predicting Equity Crowdfunding Success

Equity crowdfunding has grown exponentially in the United States since the passage of the JOBS Act in 2013, yet it continues to be a research area that is relatively unexplored in the United States due to the limited availability of data. U.S equity crowdfunding campaigns are notoriously unsuccessful, and this paper develops a predictive model for equity crowdfunding success to determine whether the positivity of the language used, and the length of the campaign description influences an investor’s decision to invest. A model is developed on a balanced training set and applied to a test set, and the overall results are evaluated using a confusion matrix to determine the accuracy, precision, and recall of the model. Our overall results indicate that both the tone of sentiment and the length of the campaign description are predictive of an equity crowdfunding campaign’s success. Specifically, the potential investors appear to be attracted to positive campaign descriptions that are written with concise language.

Prediction of cervix cancer stage and grade from diffusion weighted imaging using EfficientNet

Purpose. This study aims to introduce an innovative noninvasive method that leverages a single image for both grading and staging prediction. The grade and the stage of cervix cancer (CC) are determined from diffusion-weighted imaging (DWI) in particular apparent diffusion coefficient (ADC) maps using deep convolutional neural networks (DCNN). Methods. datasets composed of 85 patients having annotated tumor stage (I, II, III, and IV), out of this, 66 were with grade (II and III) and the remaining patients with no reported grade were retrospectively collected. The study was IRB approved. For each patient, sagittal and axial slices containing the gross tumor volume (GTV) were extracted from ADC maps. These were computed using the mono exponential model from diffusion weighted images (b-values = 0, 100, 1000) that were acquired prior to radiotherapy treatment. Balanced training sets were created using the Synthetic Minority Oversampling Technique (SMOTE) and fed to the DCNN. EfficientNetB0 and EfficientNetB3 were transferred from the ImageNet application to binary and four-class classification tasks. Five-fold stratified cross validation was performed for the assessment of the networks. Multiple evaluation metrics were computed including the area under the receiver operating characteristic curve (AUC). Comparisons with Resnet50, Xception, and radiomic analysis were performed. Results. for grade prediction, EfficientNetB3 gave the best performance with AUC = 0.924. For stage prediction, EfficientNetB0 was the best with AUC = 0.931. The difference between both models was, however, small and not statistically significant EfficientNetB0-B3 outperformed ResNet50 (AUC = 0.71) and Xception (AUC = 0.89) in stage prediction, and demonstrated comparable results in grade classification, where AUCs of 0.89 and 0.90 were achieved by ResNet50 and Xception, respectively. DCNN outperformed radiomic analysis that gave AUC = 0.67 (grade) and AUC = 0.66 (stage). Conclusion. the prediction of CC grade and stage from ADC maps is feasible by adapting EfficientNet approaches to the medical context.

Multiple adaptive over-sampling for imbalanced data evidential classification

Over-sampling approaches focus on generating samples to balance the dataset and have been widely applied in classifying imbalanced data. However, existing approaches do not take into account the uncertainty of generated samples, which may alter the data distribution and introduce uncertain information into the classification process. To tackle this issue, we propose a multiple adaptive over-sampling approach (MAO) for classifying imbalanced data based on evidence reasoning. First, we construct balanced training sets through multiple adaptive over-sampling for the minority class, which characterizes the uncertainty of over-sampling. Then, we define the intra- and inter-class inconsistency of data distribution after over-sampling to quantify the weights of different classifiers trained by various balanced subsets, weakening the negative impact of changes in data distribution on classification. Finally, we employ neighbor information to revise the results of samples that are hard to classify correctly, to avoid the risk of misclassification caused by uncertain synthetic samples to some extent. The effectiveness of MAO has been verified on several real imbalanced datasets by comparing it with other related approaches.

Integrated algorithm evaluation of a scalable digital power management innovation model

Maintaining contemporary society’s steady progress depends on the reliable functioning of the electrical grid. The complexity and nonlinearity of today’s power networks make it difficult to manage them using conventional mathematical modeling approaches. To get around these problems, researchers have started using machine learning methods, such transient stability evaluation that is based on machine learning. This research suggests a way to evaluate the transient stability of power systems by making use of the complex feature extraction capabilities of multi-layer perceptron (MLP) networks. Another approach is to suggest an ensemble learning model that uses the easy-ensemble undersampling technique to deal with power system transient stability data that is unbalanced. This method preserves unstable power system samples, performs multiple undersampling on stable samples and combines unstable and undersampling samples into multiple new balanced training sets for MLP model training. The final ensemble model is obtained through voting integration strategy. Through conducting a comparative evaluation of individual models and ensemble models, this paper discovers that the ensemble learning models exhibit superior sensitivity and accuracy in effectively addressing system instability.

Learning to Generate an Unbiased Scene Graph by Using Attribute-Guided Predicate Features

Scene Graph Generation (SGG) aims to capture the semantic information in an image and build a structured representation, which facilitates downstream tasks. The current challenge in SGG is to tackle the biased predictions caused by the long-tailed distribution of predicates. Since multiple predicates in SGG are coupled in an image, existing data re-balancing methods cannot completely balance the head and tail predicates. In this work, a decoupled learning framework is proposed for unbiased scene graph generation by using attribute-guided predicate features to construct a balanced training set. Specifically, the predicate recognition is decoupled into Predicate Feature Representation Learning (PFRL) and predicate classifier training with a class-balanced predicate feature set, which is constructed by our proposed Attribute-guided Predicate Feature Generation (A-PFG) model. In the A-PFG model, we first define the class labels of and corresponding visual feature as attributes to describe a predicate. Then the predicate feature and the attribute embedding are mapped into a shared hidden space by a dual Variational Auto-encoder (VAE), and finally the synthetic predicate features are forced to learn the contextual information in the attributes via cross reconstruction and distribution alignment. To demonstrate the effectiveness of our proposed method, our decoupled learning framework and A-PFG model are applied to various SGG models. The empirical results show that our method is substantially improved on all benchmarks and achieves new state-of-the-art performance for unbiased scene graph generation. Our code is available at https://github.com/wanglei0618/A-PFG.

A novel anomaly detection approach based on ensemble semi-supervised active learning (ADESSA)

As an industrial infrastructure, the safety and reliability of the Cyber-Physical System requires the effective anomaly detection. However, the existing detection methods have bottleneck in the face of insufficient training datasets. This work proposed and a novel anomaly detection approach based on ensemble semi-supervised active learning, which can effectively detect anomalous traffic when there is few labeled samples and the dataset is unbalanced. Specifically, this work proposed balanced sampling strategy, which combines the margin sampling and the democratic co-learning techniques, to construct a balanced training set that consists of manually labeled high-information samples and automatically labeled high-confidence samples, to effectively train the detection model on a limited budget. We also found adding correctly labeled high-confidence samples into training set improves the performance of detection model when the training samples are few and the label budget is limited. This work achieves a good balance between the effectiveness of model training and the cost of sample querying when the traffic data in CPS is rare labeled and imbalanced. In addition, we designed five pairs of experiments with NSL-KDD and SWaT dataset, and the results demonstrate the capability and advancement of proposed approach.

Balanced incremental deep reinforcement learning based on variational autoencoder data augmentation for customer credit scoring

Incremental deep reinforcement learning has been successfully applied to different fields in the real-world. However, training deep reinforcement learning models with the data set generated in a certain stage may lead to catastrophic forgetting of trained models, that is, when training models on the new data set, the performance of models will be seriously degraded on the old data set. Therefore, we construct the balanced incremental deep Q-network based on variational autoencoder data augmentation (BIDQN-VADA) model for customer credit scoring. First, the original training set is processed by the random undersampling method to obtain the balanced training set, in which the initial deep Q-network model is trained. Then, the balanced training subset with a fixed number of samples is selected from the original training set, and these samples are augmented by the variational autoencoder data augmentation technology to obtain the balanced augmented training subset. Finally, the balanced training subset and balanced augmented training subset are stored in data stream cache according to the first-in first-out principle for incrementally updating parameters of the deep Q-network model. In order to verify the performance of the model, we conduct a series of experiments on eight real-world customer credit scoring data sets. The experimental results show that the BIDQN-VADA model can achieve the best customer credit scoring performance by combining the balance, incremental and data augmentation process at the same time. More importantly, the performance of the BIDQN-VADA model is significantly better than the other seven classification models.

An omnidirectional approach to touch-based continuous authentication

This paper focuses on how touch interactions on smartphones can provide a continuous user authentication service through behaviour captured by a touchscreen. While efforts are made to advance touch-based behavioural authentication, researchers often focus on gathering data, tuning classifiers, and enhancing performance by evaluating touch interactions in a sequence rather than independently. However, such systems only work by providing data representing distinct behavioural traits. The typical approach separates behaviour into touch directions and creates multiple user profiles. This work presents an omnidirectional approach which outperforms the traditional method independent of the touch direction - depending on optimal behavioural features and a balanced training set. Thus, we evaluate five behavioural feature sets using the conventional approach against our direction-agnostic method while testing several classifiers, including an Extra-Tree and Gradient Boosting Classifier, which is often overlooked. Results show that in comparison with the traditional, an Extra-Trees classifier and the proposed approach are superior when combining strokes. However, the performance depends on the applied feature set. We find that the TouchAlytics feature set outperforms others when using our approach when combining three or more strokes. Finally, we highlight the importance of reporting the mean area under the curve and equal error rate for single-stroke performance and varying the sequence of strokes separately.

BADM-Net: Hierarchical Classification Network for Identifying Anomalous Trends in Bridge Monitoring Data Patterns

SHM systems have been widely implemented in long-span bridges, and seas of field measurement data have been accumulated. Due to the imperfect sensors, data transmission and acquisition, various anomalies inevitably exist in the SHM data, which may lead to unreliable structural condition assessment. Thus, an effective approach for detecting data anomalies is highly desirable. Due to the imbalanced data, some anomalous patterns are undertrained in popular end-to-end deep neural network models, resulting in a reduction in detection precision. In this paper, a hierarchical classification model with deep neural network tree is proposed for imbalanced data. The DNN tree contains three levels: (1) CNN to divide seven types of data into four categories (134, 2, 5, 67), denoted as C4; (2) two DNNs to classify to two classes separately (1, 34, 6, 7), denoted as D2D2; (3) DNNs to classify to two classes (3, 4). So, the DNN tree is presented as C4_D2D2_D2. The DNN tree is an open framework and can be defined based on the data characteristics. In the data processing, three data sets are built for training, namely single-channel data set, dual-channel data set and statistical data set. To validate our work, we considered the effects of balanced and imbalanced training sets and training ratios. The results show that our model can detect the multi-pattern anomalies of SHM data efficiently with 95.5% high accuracy. Besides, the proportion of abnormal data classified to normal data has been reduced, especially 3-minor. This model successfully solves the problem in a simple and easy to understand way, which has certain reference significance for the bridge structure anomaly judgment in the future.

A Diffusion Model Based on Network Intrusion Detection Method for Industrial Cyber-Physical Systems.

Industrial Cyber-Physical Systems (ICPS) connect intelligent manufacturing equipment equipped with sensors, wireless and RFID communication technologies through data interaction, which makes the interior of the factory, even between factories, become a whole. However, intelligent factories will suffer information leakage and equipment damage when being attacked by ICPS intrusion. Therefore, the network security of ICPS cannot be ignored, and researchers have conducted in-depth research on network intrusion detection for ICPS. Though machine learning and deep learning methods are often used for network intrusion detection, the problem of data imbalance can cause the model to pay attention to the misclassification cost of the prevalent class, but ignore that of the rare class, which seriously affects the classification performance of network intrusion detection models. Considering the powerful generative power of the diffusion model, we propose an ICPS Intrusion Detection system based on the Diffusion model (IDD). Firstly, data corresponding to the rare class is generated by the diffusion model, which makes the training dataset of different classes balanced. Then, the improved BiLSTM classification network is trained on the balanced training set. Extensive experiments are conducted to show that the IDD method outperforms the existing baseline method on several available datasets.

Prediction of rock mass class ahead of TBM excavation face by ML and DL algorithms with Bayesian TPE optimization and SHAP feature analysis

In this paper, field construction data from the Singapore Metro Line project were used to study the mapping relationship and establish the prediction model between TBM operation data and the ground condition ahead of the excavation face. The study presents a multi-classifier competition mechanism to construct ten separate classifiers, including logistic regression, support vector machine, random forest, extremely randomized trees, adaptive boosting machine, extreme gradient boosting (Xgboost), light gradient boosting (LightGBM), categorical boosting, long short-term memory and convolutional neural network. The acquired data were used to select 28 key TBM operating parameters by a correlation-based feature selection method, and the selected parameters in the stabilization phase after removing the outliers were calculated as the input to the classifier, and a relatively balanced training set was obtained by the synthetic minority oversampling technique. The hyperparameters of each classifier were optimized using tree Parzen estimator Bayesian optimization. The prediction results show that LightGBM presents the best results among ten different machine and deep learning algorithms with an accuracy of 96.22%, precision of 96.94%, recall of 97.33% and F1-score of 97.33%. In addition, the effect of the input parameters of the LightGBM model on the prediction accuracy of the model was analyzed using Shapley additive explanations, and the effect of sample imbalance on the prediction performance was discussed.

Deleterious synonymous mutation identification based on selective ensemble strategy.

Although previous studies have revealed that synonymous mutations contribute to various human diseases, distinguishing deleterious synonymous mutations from benign ones is still a challenge in medical genomics. Recently, computational tools have been introduced to predict the harmfulness of synonymous mutations. However, most of these computational tools rely on balanced training sets without considering abundant negative samples that could result in deficient performance. In this study, we propose a computational model that uses a selective ensemble to predict deleterious synonymous mutations (seDSM). We construct several candidate base classifiers for the ensemble using balanced training subsets randomly sampled from the imbalanced benchmark training sets. The diversity measures of the base classifiers are calculated by the pairwise diversity metrics, and the classifiers with the highest diversities are selected for integration using soft voting for synonymous mutation prediction. We also design two strategies for filling in missing values in the imbalanced dataset and constructing models using different pairwise diversity metrics. The experimental results show that a selective ensemble based on double fault with the ensemble strategy EKNNI for filling in missing values is the most effective scheme. Finally, using 40-dimensional biology features, we propose a novel model based on a selective ensemble for predicting deleterious synonymous mutations (seDSM). seDSM outperformed other state-of-the-art methods on the independent test sets according to multiple evaluation indicators, indicating that it has an outstanding predictive performance for deleterious synonymous mutations. We hope that seDSM will be useful for studying deleterious synonymous mutations and advancing our understanding of synonymous mutations. The source code of seDSM is freely accessible at https://github.com/xialab-ahu/seDSM.git.

Evaluation of four machine learning models for signal detection.

Logistic regression-based signal detection algorithms have benefits over disproportionality analysis due to their ability to handle potential confounders and masking factors. Feature exploration and developing alternative machine learning algorithms can further strengthen signal detection. Our objective was to compare the signal detection performance of logistic regression, gradient-boosted trees, random forest and support vector machine models utilizing Food and Drug Administration adverse event reporting system data. Cross-sectional study. The quarterly data extract files from 1 October 2017 through 31 December 2020 were downloaded. Due to an imbalanced outcome, two training sets were used: one stratified on the outcome variable and another using Synthetic Minority Oversampling Technique (SMOTE). A crude model and a model with tuned hyperparameters were developed for each algorithm. Model performance was compared against a reference set using accuracy, precision, F1 score, recall, the receiver operating characteristic area under the curve (ROCAUC), and the precision-recall curve area under the curve (PRCAUC). Models trained on the balanced training set had higher accuracy, F1 score and recall compared to models trained on the SMOTE training set. When using the balanced training set, logistic regression, gradient-boosted trees, random forest and support vector machine models obtained similar performance evaluation metrics. The gradient-boosted trees hyperparameter tuned model had the highest ROCAUC (0.646) and the random forest crude model had the highest PRCAUC (0.839) when using the balanced training set. All models trained on the balanced training set performed similarly. Logistic regression models had higher accuracy, precision and recall. Logistic regression, random forest and gradient-boosted trees hyperparameter tuned models had a PRCAUC ⩾ 0.8. All models had an ROCAUC ⩾ 0.5. Including both disproportionality analysis results and additional case report information in models resulted in higher performance evaluation metrics than disproportionality analysis alone.

Probabilistic learning for pulsar classification

In this work, we explore the possibility of using probabilistic learning to identify pulsar candidates. We make use of Deep Gaussian Process (DGP) and Deep Kernel Learning (DKL). Trained on a balanced training set in order to avoid the effect of class imbalance, the performance of the models, achieving relatively high probability of differentiating the positive class from the negative one (roc-auc ∼ 0.98), is very promising overall. We estimate the predictive entropy of each model predictions and find that DKL is more confident than DGP in its predictions and provides better uncertainty calibration. Upon investigating the effect of training with imbalanced dataset on the models, results show that each model performance decreases with an increasing number of the majority class in the training set. Interestingly, with a number of negative class 10× that of positive class, the models still provide reasonably well calibrated uncertainty, i.e. an expected Uncertainty Calibration Error (UCE) less than 6%. We also show in this study how, in the case of relatively small amount of training dataset, a convolutional neural network based classifier trained via Bayesian Active Learning by Disagreement (BALD) performs. We find that, with an optimized number of training examples, the model — being the most confident in its predictions — generalizes relatively well and produces the best uncertainty calibration which corresponds to UCE = 3.118%.

Expensive Multiobjective Optimization by Relation Learning and Prediction

Expensive multiobjective optimization problems pose great challenges to evolutionary algorithms due to their costly evaluation. Building cheap surrogate models to replace the expensive real models has been proved to be a practical way to reduce the number of costly evaluations. Supervised learning techniques from the community of machine learning have been widely applied to build either regressors, which approximate the fitness values of candidate solutions, or classifiers, which estimate the categories of candidate solutions. Considering the characteristics of the data produced in optimization, this article proposes a new surrogate model, called a relation model, for evolutionary multiobjective optimization. Instead of estimating the qualities of candidate solutions directly, the relation model tries to estimate the relationship between a pair of solutions <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\langle \mathbf {x}, \mathbf {y}\rangle $ </tex-math></inline-formula> , i.e., <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\mathbf {x}$ </tex-math></inline-formula> dominates <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\mathbf {y}$ </tex-math></inline-formula> , <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\mathbf {x}$ </tex-math></inline-formula> is dominated by <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\mathbf {y}$ </tex-math></inline-formula> , or <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\mathbf {x}$ </tex-math></inline-formula> is nondominated with <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\mathbf {y}$ </tex-math></inline-formula> in the case of multiobjective optimization. To implement this idea, first a balanced training set is prepared, then a classifier is built based on the training data set to learn the relationship, and finally, the classifier with a voting-scoring strategy is applied to estimate the relationship between the candidate solutions and parent solutions. By this way, the promising candidate solutions are recognized and evaluated by the real models. The new approach is applied to three well-known benchmark suites and two real-world applications, and the experimental results suggest that the proposed method outperforms some state-of-the-art methods based on regression and classification models on the given instances.

Evaluation of deep learning and transform domain feature extraction techniques for land cover classification: balancing through augmentation.

The identification of features that can improve classification accuracy is a major concern in land cover classification research. This paper compares deep learning and transform domain feature extraction techniques for land cover classification of SAR data on balanced and imbalanced training sets. Convolutional autoencoders (CAE), variational autoencoders (VAE), and Haar wavelet transforms (HWT) are used and evaluated for feature generation capability. Variations in features of CAE and HWT help gather more information about the image patch. The fusion of CAE and HWT features provides a combination of high- and low-frequency coefficients, respectively, which improves classification accuracy for various land covers. To assess features generated through fusion of features, RISAT-1 C-band and AIRSAR L-band datasets are used. Furthermore, agricultural/grass land has similar features with open forest which leads to misclassification of forest in agricultural/grass land. Increasing the number of samples in each class using the Synthetic Minority Oversampling Technique (SMOTE) increases training samples. Hierarchical classification of the above-mentioned features, where agricultural/grass land and forest classes are discriminated, improves classification results. This paper evaluates all the three types of features and fused features and provides a guidance for land cover classification.

Machine learning for maternal health: Predicting delivery location in a community health worker program in Zanzibar.

BackgroundMaternal and neonatal health outcomes in low- and middle-income countries (LMICs) have improved over the last two decades. However, many pregnant women still deliver at home, which increases the health risks for both the mother and the child. Community health worker programs have been broadly employed in LMICs to connect women to antenatal care and delivery locations. More recently, employment of digital tools in maternal health programs have resulted in better care delivery and served as a routine mode of data collection. Despite the availability of rich, patient-level data within these digital tools, there has been limited utilization of this type of data to inform program delivery in LMICs.MethodsWe use program data from 38,787 women enrolled in Safer Deliveries, a community health worker program in Zanzibar, to build a generalizable prediction model that accurately predicts whether a newly enrolled pregnant woman will deliver in a health facility. We use information collected during the enrollment visit, including demographic data, health characteristics and current pregnancy information. We apply four machine learning methods: logistic regression, LASSO regularized logistic regression, random forest and an artificial neural network; and three sampling techniques to address the imbalanced data: undersampling of facility deliveries, oversampling of home deliveries and addition of synthetic home deliveries using SMOTE.ResultsOur models correctly predicted the delivery location for 68%–77% of the women in the test set, with slightly higher accuracy when predicting facility delivery versus home delivery. A random forest model with a balanced training set created using undersampling of existing facility deliveries accurately identified 74.4% of women delivering at home.ConclusionsThis model can provide a “real-time” prediction of the delivery location for new maternal health program enrollees and may enable early provision of extra support for individuals at risk of not delivering in a health facility, which has potential to improve health outcomes for both mothers and their newborns. The framework presented here is applicable in other contexts and the selection of input features can easily be adapted to match data availability and other outcomes, both within and beyond maternal health.

Research on Improved Random Forest Algorithm for Highly Unbalanced Data

To address the problems of low prediction performance of minority class samples due to highly unbalanced data and improved prediction performance of minority class but decreased prediction performance of majority class after balanced training set, an improved algorithm combining feature fusion and random forests quantile classifier is proposed. The algorithm obtains multiple data subsets by feature segmentation through sliding windows, and then inputs the obtained data subsets into the quantile random forest for cross-validation to obtain multiple class probability vectors for each sample, and finally splices the obtained class probability vectors with the original data set as new features to form a new data set, and uses the new combination of features to better represent the relationship between minority and majority classes, thus improve the overall classification performance of the classifier. Experiments on the KEEL dataset show that the improved algorithm improves the overall sample classification performance for unbalanced data compared to other algorithms.