Supervised Machine Learning Process Research Articles

A Typology of Social Media Use by Human Service Nonprofits: Mixed Methods Study.

Nonprofit organizations are increasingly using social media to improve their communication strategies with the broader population. However, within the domain of human service nonprofits, there is hesitancy to fully use social media tools, and there is limited scope among organizational personnel in applying their potential beyond self-promotion and service advertisement. There is a pressing need for greater conceptual clarity to support education and training on the varied reasons for using social media to increase organizational outcomes. This study leverages the potential of Twitter (subsequently rebranded as X [X Corp]) to examine the online communication content within a sample (n=133) of nonprofit sexual assault (SA) centers in Canada. To achieve this, we developed a typology using a qualitative and supervised machine learning model for the automatic classification of tweets posted by these centers. Using a mixed methods approach that combines machine learning and qualitative analysis, we manually coded 10,809 tweets from 133 SA centers in Canada, spanning the period from March 2009 to March 2023. These manually labeled tweets were used as the training data set for the supervised machine learning process, which allowed us to classify 286,551 organizational tweets. The classification model based on supervised machine learning yielded satisfactory results, prompting the use of unsupervised machine learning to classify the topics within each thematic category and identify latent topics. The qualitative thematic analysis, in combination with topic modeling, provided a contextual understanding of each theme. Sentiment analysis was conducted to reveal the emotions conveyed in the tweets. We conducted validation of the model with 2 independent data sets. Manual annotation of 10,809 tweets identified seven thematic categories: (1) community engagement, (2) organization administration, (3) public awareness, (4) political advocacy, (5) support for others, (6) partnerships, and (7) appreciation. Organization administration was the most frequent segment, and political advocacy and partnerships were the smallest segments. The supervised machine learning model achieved an accuracy of 63.4% in classifying tweets. The sentiment analysis revealed a prevalence of neutral sentiment across all categories. The emotion analysis indicated that fear was predominant, whereas joy was associated with the partnership and appreciation tweets. Topic modeling identified distinct themes within each category, providing valuable insights into the prevalent discussions surrounding SA and related issues. This research contributes an original theoretical model that sheds light on how human service nonprofits use social media to achieve their online organizational communication objectives across 7 thematic categories. The study advances our comprehension of social media use by nonprofits, presenting a comprehensive typology that captures the diverse communication objectives and contents of these organizations, which provide content to expand training and education for nonprofit leaders to connect and engage with the public, policy experts, other organizations, and potential service users.

Frequency Domain Analysis of Endocardial Electrograms for Detection of Nontransmural Myocardial Fibrosis in Nonischemic Cardiomyopathy

BackgroundVoltage mapping in nonischemic cardiomyopathy can fail to identify midmyocardial substrate for ventricular arrhythmias, an important cause of ablation failure. ObjectivesThe aim of this study was to assess whether frequency domain analysis of endocardial left ventricular electrograms (EGMs) can better predict the presence of midmyocardial fibrosis (MMF) compared with voltage amplitude. MethodsNonischemic cardiomyopathy patients undergoing ventricular tachycardia ablation with registered preprocedural cardiac computed tomography and late iodine enhancement were included. Presence of fibrosis at each EGM site was assessed. Bipolar and unipolar EGMs were transformed to the frequency domain using multitaper spectral analysis. Singular value decomposition of the EGM frequency spectrum was used within a supervised machine learning process to select features to predict the presence of MMF and compare against predictions using voltage amplitude. ResultsThirteen patients were included (median age 57 years [IQR: 28-73 years], median ejection fraction 40% [IQR: 15%-57%]). A total of 6,015 EGM pairs were processed: 2,459 EGM pairs in MMF areas and 3,556 EGM pairs in non-MMF areas. Supervised classifiers were trained with stratified k-fold cross-validation within patients. The distribution of mean area under the curve metrics using frequency features, f, was significantly greater than voltage feature area under the curve metrics, v, (mean f = 0.841 [95% CI: 0.789-0.884] vs mean v = 0.591 [95% CI: 0.530-0.658]; P < 0.001), indicating that frequency-trained classifiers better predicted the presence of MMF. ConclusionsThese data indicate the promising discriminatory value of endocardial EGM frequency content in the assessment of concealed myocardial substrate. Further studies are needed to investigate the importance of the specific frequency features identified.

A New Approach for Assessing Heat Balance State along a Water Transfer Channel during Winter Periods

Ice problems in channels for water transfer in cold regions seriously affect the capacity and efficiency of water conveyance. Sometimes, ice problems such as ice jams in water transfer channels create risk during winter periods. Recently, water temperature and environmental factors at various cross-sections along the main channel of the middle route of the South-to-North Water Transfer Project in China have been measured. Based on these temperature data, the heat balance state of this water transfer channel has been investigated. A principal component analysis (PCA) method has been used to analyze the complex factors influencing the observed variations of the water temperature, by reducing eigenvector dimension and then extracting the principal component as the input feature. Based on the support vector machine (SVM) theory, a new approach for judging the heat loss or heat gain of flowing water in a channel during winter periods has been developed. The Gaussian radial basis is used as the kernel function in this new approach. Then, parameters have been optimized by means of various methods. Through the supervised machine learning process toward the observed water temperature data, it is found that the air–water temperature difference and thermal conditions are the key factors affecting the heat loss or heat absorption of water body. Results using the proposed method agree well with those of measurements. The changes of water temperature are well predicted using the proposed method together with the state of water heat balance.

Supervised Classification of Motor-Rehabilitation Body Movements with RGB Cameras and Pose Tracking Data

The technological evolution allowed the use of a single camera for precise and effective body tracking, reducing the cost and increasing the accessibility of applications in places where depth cameras and wearable sensors are not available. This paper describes and implements a supervised machine learning process consisting of a mobile application used as a motion capture device which also transforms the data into an input for a machine learning model that classifies upper and lower limbs movements (24 types of human movements). The user performs movements in front of the camera, and the trained model classifies them. We designed the system to work in a motor-rehabilitation context to assist the professional while the patient does physical exercises. The implementation can summarize the movements made during the rehabilitation sessions by counting the repetitions and classifying them when done completely or reached a specific range of motion.

Development of predictive modeling and deep learning classification of taxi trip tolls

Several studies discussed the predictive modeling of deep learning in different applications such as classifying tissue features from microstructural data, Crude Oil Prices, mechanical constitutive behavior of materials, microbiome data, and mineral prospectively. Commercial navigation includes a wealth of trip-related data, including distance, expected journey time, and tolls that may be encountered along the way. Using a classification algorithm, it is possible to extract drop-off and pickup locations from taxi trip data and estimate if the tour would incur tolls. In this work, let’s use the classification learner to create classification models, compare their performance, and export the findings for additional study. The workflow for the classification learner is the same as for the regression learner. The purpose is to make predictions based on fresh data in order to see how well the model performs with new data. To train the model, it’s critical to separate the data set. The combined training and validation data is next pre-processed, which involves tasks such as cleaning and developing new features skills. Once the data has been prepared, it’s time to begin the supervised machine learning process and test a number of ways to identify the best model, such as the type of model that should be used, the important features, and the best parameters of the model to find the best fit for the considered data. The results of analyzing different predictive multiclass classification models with taxi trip tolls show that it is possible to use a machine learning-based model when we like to avoid road tolls depending on historical data on taxi trip tolls. The outcome of this study can help to expect road tolls from the drop-off and pickup locations of a taxi data

Young drivers’ pedestrian anti-collision braking operation data modelling for ADAS development

Smart cities and smart mobility come from intelligent systems designed by humans. Artificial Intelligence (AI) is contributing significantly to the development of these systems, and the automotive industry is the most prominent example of "smart" technology entering the market: there are Advanced Driver Assistance System (ADAS), Radar/LIDAR detection units and camera-based Computer Vision systems that can assess driving conditions. Actually, these technologies have become consumer goods and services in mass-produced vehicles to provide human drivers with tools for a more comfortable and safer driving. Nevertheless, they need to be further improved for progress in the transition to fully automated driving or simply to increase vehicle automation levels. To this end, it becomes imperative to accurately predict driver’s decisions, model human driving behaviors, and introduce more accurate risk assessment metrics. This paper presents a system that can learn to predict the future braking behavior of a driver in a typically urban vehicle-pedestrian conflict, i.e., when a pedestrian enters a zebra crossing from the curb and a vehicle is approaching. The algorithm proposes a sequential prediction of relevant operational indicators that continuously describe the encounter process. A car driving simulator was used to collect reliable data on braking behaviours of a cohort of 68 licensed university students, who faced the same urban scenario. The vehicle speed, steering wheel angle, and pedal activity were recorded as the participants approached the crosswalk, along with the azimuth angle of the pedestrian and the relative longitudinal distance between the vehicle and the pedestrian: the proposed system employs the vehicle information as human driving decisions and the pedestrian information as explanatory variables of the environmental state. In fact, the pedestrian’s polar coordinates are usually calculated by an on-board millimeter-wave radar which is typically used to perceive the environment around a vehicle. All mentioned information is represented in the form of time series data and is used to train a recurrent neural network in a supervised machine learning process. The main purpose of this research is to define a system of behavioral profiles in non-collision conditions that could be used for enhancing the existing intelligent driving systems, e.g., to reduce the number of warnings when the driver is not on a collision course with a pedestrian. Preliminary experiments reveal the feasibility of the proposed system.

Classification of the Energy Production Potential of Water Lens Solar Concentrators Using Machine Learning

Assessing the potential of renewable energy sources for buildings in neighborhoods becomes a crucial task in the early planning stage. Integrating solar energy equipment into urban buildings poses many challenges, such as uncertainties and the complexity of urban built agglomeration. Due to the time-consuming solar energy potential assessment process and lack of knowledge of urban actors, a reliable framework is required to predict buildings’ solar energy potential. This research presents a comprehensive machine learning data processing framework to predict output energy of Water Lenses (WL) based on buildings specifications and relationship to the neighbourhood. The research used a raw dataset consisting of 7000 sample buildings in different situations by applying 12 years of climatic conditions in Tallinn, Estonia. The results were entered into a Supervised Machine Learning process and the Gaussian Naive Bayes technique was used for classification of building features to be implemented with solar systems. Finally, the process was measured by a confusion matrix that showed 80% accuracy of ML output predictions in the urban context.

Entropy-Based Time Window Features Extraction for Machine Learning to Predict Acute Kidney Injury in ICU

Acute kidney injury (AKI) refers to rapid decline of kidney function and is manifested by decreasing urine output or abnormal blood test (elevated serum creatinine). Electronic health records (EHRs) is fundamental for clinicians and machine learning algorithms to predict the clinical outcome of patients in the Intensive Care Unit (ICU). Early prediction of AKI could automatically warn the clinicians to review the possible risk factors and act in advance to prevent it. However, the enormous amount of patient data usually consists of a relatively incomplete data set and is very challenging for supervised machine learning process. In this paper, we propose an entropy-based feature engineering framework for vital signs based on their frequency of records. In particular, we address the missing at random (MAR) and missing not at random (MNAR) types of missing data according to different clinical scenarios. Regarding its applicability, we applied it to establish a prediction model for future AKI in ICU patients using 4278 ICU admissions from a tertiary hospital. Our result shows that the proposed entropy-based features are feasible to be used in the AKI prediction model and its performance improves as the data availability increases. In addition, we study the performance of AKI prediction model by comparing different time gaps and feature windows with the proposed vital sign entropy features. This work could be used as a guidance for feature windows selection and missing data processing during the development of a prediction model in ICU.

Artificial Intelligence and Robotics in Regional Anaesthesia: Do they have a role?

Artificial Intelligence and Robotics in Regional Anaesthesia: Do they have a role?

Algorithmically generated malicious domain names detection based on n-grams features

Botnets are one of the major cyber infections used in several criminal activities. In most botnets, a Domain Generation Algorithm (DGA) is used by bots to make DNS queries aimed at establishing the connection with the Command and Control (C&C) server. The identification of such queries by monitoring the network DNS traffic is then crucial for bot detection. In this paper we present a methodology to detect DGA generated domain names based on a supervised machine learning process, trained with a dataset of known benign and malicious domain names. The proposed approach represents the domain names through a set of features which express the similarity between the 2-grams and 3-grams in a single unclassified domain name and those in domain names known as malicious or benign. We used the Kullback-Leibner divergence and the Jaccard Index to estimate the similarity, and we tested different machine learning algorithms to classify each domain name as benign or DGA-based (with both binary and multi-class approach). The results of our experiments demonstrate that the proposed methodology, which only exploits lexical features of domain names, attains a good level of accuracy and results in a general model able to classify previously unseen domains in an effective way. It is also able to outperform some of the state-of-the-art featur eless classification methods based on deep learning.

Achalasia subtypes can be identified with functional luminal imaging probe (FLIP) panometry using a supervised machine learning process.

Achalasia subtypes on high-resolution manometry (HRM) prognosticate treatment response and help direct management plan. We aimed to utilize parameters of distension-induced contractility and pressurization on functional luminal imaging probe (FLIP) panometry and machine learning to predict HRM achalasia subtypes. One hundred eighty adult patients with treatment-naïve achalasia defined by HRM per Chicago Classification (40 type I, 99 type II, 41 type III achalasia) who underwent FLIP panometry were included: 140 patients were used as the training cohort and 40 patients as the test cohort. FLIP panometry studies performed with 16-cm FLIP assemblies were retrospectively analyzed to assess distensive pressure and distension-induced esophageal contractility. Correlation analysis, single tree, and random forest were adopted to develop classification trees to identify achalasia subtypes. Intra-balloon pressure at 60mL fill volume, and proportions of patients with absent contractile response, repetitive retrograde contractile pattern, occluding contractions, sustained occluding contractions (SOC), contraction-associated pressure changes >10mmHg all differed between HRM achalasia subtypes and were used to build the decision tree-based classification model. The model identified spastic (type III) vs non-spastic (types I and II) achalasia with 90% and 78% accuracy in the train and test cohorts, respectively. Achalasia subtypes I, II, and III were identified with 71% and 55% accuracy in the train and test cohorts, respectively. Using a supervised machine learning process, a preliminary model was developed that distinguished type III achalasia from non-spastic achalasia with FLIP panometry. Further refinement of the measurements and more experience (data) may improve its ability for clinically relevant application.

WIC Participation and Relative Quality of Household Food Purchases: Evidence from FoodAPS

We examine the effect of the Special Supplemental Nutrition Program for Women, Infants, and Children (WIC) on the quality of household food purchases using the National Household Food Acquisition and Purchase Survey (FoodAPS) and propensity score matching. A healthy purchasing index (HPI) is used to measure nutritional quality of household food purchases. WIC foods explain the improvement in quality of food purchases, not self‐selection of more nutrition‐conscious households into the program. The improvement in purchase quality was driven entirely by WIC participating households who redeemed WIC foods during the interview week. There was no significant difference between WIC participants who did not redeem WIC foods and eligible nonparticipants. In this sample, there is no evidence that lack of access to clinics has adverse effects on participation nor is there evidence that HPI depends on supermarket access. A supervised machine learning process supports our main conclusion on the importance of WIC foods.

Systematic Labeling Bias in Galaxy Morphologies

Abstract We present a metric to quantify systematic labeling bias in galaxy morphology data sets stemming from the quality of the labeled data. This labeling bias is independent from labeling errors and requires knowledge about the intrinsic properties of the data with respect to the observed properties. We conduct a relative comparison of label bias for different low-redshift galaxy morphology data sets. We show our metric is able to recover previous de-biasing procedures based on redshift as biasing parameter. By using the image resolution instead, we find biases that have not been addressed. We find that the morphologies based on supervised machine learning trained over features such as colors, shape, and concentration show significantly less bias than morphologies based on expert or citizen-science classifiers. This result holds even when there is underlying bias present in the training sets used in the supervised machine learning process. We use catalog simulations to validate our bias metric and show how to bin the multi-dimensional intrinsic and observed galaxy properties used in the bias quantification. Our approach is designed to work on any other labeled multi-dimensional data set, and the code is publicly available (https://github.com/guille-c/labeling_bias).

Syntactic parsing and supervised analysis of Sindhi text

This research study addresses the morphological and syntactic problems of Sindhi language text by proposing an Algorithm for tokenization and syntactic parsing. A Sindhi parser is developed on basis of proposed algorithm to perform syntactic parsing on Sindhi text using Sindhi WordNet (SWN) and corpus. Results of Sindhi syntactic parsing are accumulated to develop multi-class and multi-feature based Sindhi dataset in CSV format. Three attributes of Sindhi dataset are labelled as class. All three classes are comprised with different number of categories. SVM, Random forest and K-NN supervised machine learning methods are used and trained to analyze and evaluate the Sindhi dataset. 80% of dataset is used as training set and 20% of dataset is used as test set. In this research study, 10-fold cross validation technique is applied to evaluate and validate the supervised machine learning process. The SVM classifier gives better results on class phrase and UPOS whereas Random forest gives better result on class TagStatus. Precision, recall, f-measure and confusion matrix approve the performance of all supervised classifiers. The better performance of supervised machine learning methods, support the Sindhi dataset and Sindhi online parser for future research. This study opens new doors for research on right hand written languages especially Sindhi language to solve its computational linguistics problems.

Recognition of pigment network pattern in dermoscopy images based on fuzzy classification of pixels

Background and ObjectiveOne of the most relevant dermoscopic patterns is the pigment network. An innovative method of pattern recognition is presented for its detection in dermoscopy images. MethodsIt consists of two steps. In the first one, by means of a supervised machine learning process and after performing the extraction of different colour and texture features, a fuzzy classification of pixels into the three categories present in the pattern’s definition (“net”, “hole” and “other”) is carried out. This enables the three corresponding fuzzy sets to be created and, as a result, the three probability images that map them out are generated. In the second step, the pigment network pattern is characterised from a parameterisation process –derived from the system specification– and the subsequent extraction of different features calculated from the combinations of image masks extracted from the probability images, corresponding to the alpha-cuts obtained from the fuzzy sets. ResultsThe method was tested on a database of 875 images –by far the largest used in the state of the art to detect pigment network– extracted from a public Atlas of Dermoscopy, obtaining AUC results of 0.912 and 88%% accuracy, with 90.71%% sensitivity and 83.44%% specificity. ConclusionThe main contribution of this method is the very design of the algorithm, highly innovative, which could also be used to deal with other pattern recognition problems of a similar nature. Other contributions are: 1. The good performance in discriminating between the pattern and the disturbing artefacts –which means that no prior preprocessing is required in this method– and between the pattern and other dermoscopic patterns; 2. It puts forward a new methodological approach for work of this kind, introducing the system specification as a required step prior to algorithm design and development, being this specification the basis for a required parameterisation –in the form of configurable parameters (with their value ranges) and set threshold values– of the algorithm and the subsequent conducting of the experiments.

A neural network approach for fast, automated quantification of DIR performance.

A critical step in adaptive radiotherapy (ART) workflow is deformably registering the simulation CT with the daily or weekly volumetric imaging. Quantifying the deformable image registration accuracy under these circumstances is a complex task due to the lack of known ground-truth landmark correspondences between the source data and target data. Generating landmarks manually (using experts) is time-consuming, and limited by image quality and observer variability. While image similarity metrics (ISM) may be used as an alternative approach to quantify the registration error, there is a need to characterize the ISM values by developing a nonlinear cost function and translate them to physical distance measures in order to enable fast, quantitative comparison of registration performance. In this paper, we present a proof-of-concept methodology for automated quantification of DIR performance. A nonlinear cost function was developed as a combination of ISM values and governed by the following two expectations for an accurate registration: (a) the deformed data obtained from transforming the simulation CT data with the deformation vector field (DVF) should match the target image data with near perfect similarity, and (b) the similarity between the simulation CT and deformed data should match the similarity between the simulation CT and the target image data. A deep neural network (DNN) was developed that translated the cost function values to actual physical distance measure. To train the neural network, patient-specific biomechanical models of the head-and-neck anatomy were employed. The biomechanical model anatomy was systematically deformed to represent changes in patient posture and physiological regression. Volumetric source and target images with known ground-truth deformations vector fields were then generated, representing the daily or weekly imaging data. Annotated data was then fed through a supervised machine learning process, iteratively optimizing a nonlinear model able to predict the target registration error (TRE) for given ISM values. The cost function for sub-volumes enclosing critical radiotherapy structures in the head-and-neck region were computed and compared with the ground truth TRE values. When examining different combinations of registration parameters for a single DIR, the neural network was able to quantify DIR error to within a single voxel for 95% of the sub-volumes examined. In addition, correlations between the neural network predicted error and the ground-truth TRE for the Planning Target Volume and the parotid contours were consistently observed to be > 0.9. For variations in posture and tumor regression for 10 different patients, patient-specific neural networks predicted the TRE to within a single voxel > 90% on average. The formulation presented in this paper demonstrates the ability for fast, accurate quantification of registration performance. DNN provided the necessary level of abstraction to estimate a quantified TRE from the ISM expectations described above, when sufficiently trained on annotated data. In addition, biomechanical models facilitated the DNN with the required variations in the patient posture and physiological regression. With further development and validation on clinical patient data, such networks have potential impact in patient and site-specific optimization, and stream-lining clinical registration validation.

Fitness assessment of document model

Document classification is a supervised machine learning process, where predefined category labels are assigned to documents based on the hypothesis derived from training set of labelled documents. Documents cannot be directly interpreted by a computer system unless they have been modelled as a collection of computable features. Rogati and Yang [M. Rogati and Y. Yang, “Resource selection for domain-specific cross-lingual IR”, in SIGIR 2004: Proceedings of the 27th annual international conference on Research and Development in Information Retrieval, ACM Press, Sheffied: United Kingdom, pp. 154–161.] pointed out that the effectiveness of document classification system may vary in different domains. This implies that the quality of document model contributes to the effectiveness of document classification. Conventionally, model evaluation is accomplished by comparing the effectiveness scores of classifiers on model candidates. However, this kind of evaluation methods may encounter either under-fitting or over-fitting problems, because the effectiveness scores are restricted by the learning capacities of classifiers. We propose a model fitness evaluation method to determine whether a model is sufficient to distinguish positive and negative instances while still competent to provide satisfactory effectiveness with a small feature subset. Our experiments demonstrated how the fitness of models are assessed. The results of our work contribute to the researches of feature selection, dimensionality reduction and document classification.