Machine Learning Techniques For Classification Research Articles

Do Large Datasets or Hybrid Integrated Models Outperform Simple Ones in Predicting Commodity Prices and Foreign Exchange Rates?

With the continuous advancement of machine learning and the increasing availability of internet-based information, there is a belief that these approaches and datasets enhance the accuracy of price prediction. However, this study aims to investigate the validity of this claim. The study examines the effectiveness of a large dataset and sophisticated methodologies in forecasting foreign exchange rates (FX) and commodity prices. Specifically, we employ sentiment analysis to construct a robust sentiment index and explore whether combining sentiment analysis with machine learning surpasses the performance of a large dataset when predicting FX and commodity prices. Additionally, we apply machine learning methodologies such as random forest (RF), eXtreme gradient boosting (XGB), and long short-term memory (LSTM), alongside the classical statistical model autoregressive integrated moving average (ARIMA), to forecast these prices and compare the models’ performance. Based on the results, we propose novel methodologies that integrate wavelet transformation with classical ARIMA and machine learning techniques (seasonal-decomposition-ARIMA-LSTM, wavelet-ARIMA-LSTM, wavelet-ARIMA-RF, wavelet-ARIMA-XGB). We apply this analysis procedure to the commodity gold futures prices and the euro foreign exchange rates against the US dollar.

Data-driven acoustic measurement of moisture content in flowing biomass

Measuring the moisture content in flowing biomass is critical to processes such as liquid biofuel conversion, such as biogasoline, biodiesel, bio jet kerosene, etc. However, biomass tends to flow in aggregates, which results in significant inhomogeneities in the amount of biomass flowing in front of a sensor at a given time, and there can be significant overlap in the material properties of dry vs wet biomass, leading to poor signal-to-noise ratio. We present a technique for identifying biomass moisture content using a series of acoustic pitch-catch measurements to quantify the sound speed and acoustic amplitude through the biomass, in conjunction with classical machine learning techniques, including Naive Bayes, Random Forest, and K-Nearest Neighbors classification. We amplify the differences between the acoustic measurements in different moisture levels by collecting a series of pulse-echo measurements, which we sort in order of ascending sound speed. We test the accuracy of the technique on experimentally-prepared batches of corn stover biomass with specified moisture levels, and measure the average error in the estimated moisture level as a function of the number of pitch-catch measurements used. We observe average estimation errors as low as 6.7% by increasing the number of measurements and optimizing the hyperparameters. This work presents a novel method determining moisture content in flowing biomass with inhomogeneous flow. Additionally, this technique has application in optimizing biomass conversion processes, as well as other fields including, paper production, natural fiber processing, and mineral extraction.

Video-Based Recognition of Human Activity Using Novel Feature Extraction Techniques

This paper proposes a novel approach to activity recognition where videos are compressed using video coding to generate feature vectors based on compression variables. We propose to eliminate the temporal domain of feature vectors by computing the mean and standard deviation of each variable across all video frames. Thus, each video is represented by a single feature vector of 67 variables. As for the motion vectors, we eliminated their temporal domain by projecting their phases using PCA, thus representing each video by a single feature vector with a length equal to the number of frames in a video. Consequently, complex classifiers such as LSTM can be avoided and classical machine learning techniques can be used instead. Experimental results on the JHMDB dataset resulted in average classification accuracies of 68.8% and 74.2% when using the projected phases of motion vectors and video coding feature variables, respectively. The advantage of the proposed solution is the use of FVs with low dimensionality and simple machine learning techniques.

Phishing URL Detection using Machine Learning

Abstract: Phishing attacks continue to pose a major threat for computer system defenders, often forming the first step in a multistage attack. There have been great strides made in phishing detection; however,some phishing emails appear to pass through filters by making simple structural and semantic changes to the messages. We tackle this problem through the use of a machine learning classifier operating on a large corpus of phishing and legitimate emails. We design a system to extract features, elevating some to higherlevel feature, that are meant to defeat common phishing email detection strategies. This paper presents an approach to detect phishing URLs in an efficient way based on URL features only. For detecting the phishing URLs SVM classifier is used. The performances are evaluated for different size of datasets using different number of features. The results are compared with other machine learning classification techniques. The proposed system is able to detect phishing websites using URL features only.

UAV-derived photogrammetric point clouds and multispectral indices for fuel estimation in Mediterranean forests

Sensors attached to unmanned aerial vehicles (UAVs) allow estimating a large number of forest attributes related to forest fuels. This study assesses photogrammetric point clouds and multispectral indices obtained from a fixed-wing UAV for the classification of Prometheus fuel types in 82 forest plots in Aragón (NE Spain). Images captured by an RGB camera and a multispectral sensor allowed generating high density photogrammetric point clouds (RGB: 3000 points/m2; multispectral: 85 points/m2), which were normalized using alternatively a Digital Elevation Model (DEM) of 0.5, 1, and 2 m resolution. A set of structural and textural variables were derived from the normalized point cloud heights, and for the latter, the gray-level co-occurrence matrix (GLCM) approach was used. Multispectral images were also used to create seven spectral vegetation indices. The most relevant structural, textural, and spectral variables to introduce into the fuel types classification models were selected using Dunn's test, which included: the vegetation height at the 50th percentile, the coefficient of variation of the heights, the percentage of returns above 4 m, the mean textural dissimilarity, and the mean of the Green Chlorophyll Index. Three different data samples were introduced in the models: i) the relevant structural and textural variables from the RGB camera (RGB data sample); ii) the relevant structural, textural, and spectral variables from the multispectral sensor (MS data sample); and iii) the relevant structural and textural variables from the RGB camera plus the relevant spectral variable from the multispectral sensor (integrated data sample). After comparing three machine learning classification techniques (Random Forest, and Linear and Radial Support Vector Machine), the best results were obtained with Random Forest with k-fold cross-validation (k-10) and the integrated data sample with normalized point clouds at 0.5 m DEM resolution (overall accuracy = 71%). The variables successfully identified the Prometheus main fire carriers (i.e., shrubs or trees) and confusions were mainly located within the fuel types of the same dominant stratum, especially in fuel types 3 and 6. These results demonstrate the ability of UAV imagery to classify forest fuels in Mediterranean environments when RGB and multispectral data are combined.

Integration of Per- and Polyfluoroalkyl Substance (PFAS) Fingerprints in Fish with Machine Learning for PFAS Source Tracking in Surface Water

Per- and polyfluoroalkyl substances (PFASs) are a class of environmental contaminants that originate from various sources. The unique chemical fingerprints associated with many commercial products and industrial applications make PFASs ideal candidates for machine learning (ML)-assisted environmental forensics. Here, we propose a novel use of PFAS fingerprints in fish tissue from surface water systems to classify exposure from multiple sources of PFASs using a proof-of-concept demonstration. Three supervised ML classification techniques (k-nearest neighbors (KNN), decision trees, support vector machines) implementing two predictive features are used to classify literature-reported PFAS fingerprints in fish (n = 1057). The importance of additional predictive features was explored using brute force optimization of a multifeature KNN algorithm. The multiclass classification considered exposure to aqueous film-forming foam-impacted water, paper industry wastewater, diffuse sources, or PFASs undergoing long-range transport. The optimized classifiers demonstrated 85%–94% classification accuracy for this first known multiclass classification of PFASs for environmental forensics. The optimized classifiers also demonstrated 79%–92% classification accuracy with a set of independent external validation data (n = 192). Our results demonstrate that PFAS fingerprints in fish tissue may be an effective means of PFAS source tracking in surface water systems. The source code is provided for guidance on best practices for ML-assisted environmental forensics.

‘MrudaTest’: An Android Phone Intervened Platform for Soil Primary Nutrients and pH Detection

ABSTRACT ‘Mruda’ is a Marathi term that signifies soil, and ‘MrudaTest’ shows soil testing. Crop yield is influenced by soil pH and macronutrients such as nitrogen (N), phosphorus (P), and potassium (K). Precision agriculture, which has gained appeal on a global scale, can be achieved in part by managing fertilizers properly in accordance with soil health. The standard laboratory test is time-consuming and resource-intensive. Sample testing takes longer duration due to overworked soil testing labs, which discourages farmers from performing soil fertility tests. To address the aforementioned issue, portable, farmer-friendly, rapid-testing technology is required, and our effort is geared in that direction. Soil samples were treated with reagents that turn the clear solution into color as per soil nutrients. An android phone was used to detect and quantify color information using a built-in camera, computing hardware, graphical user interface, image processing, and machine learning classification techniques. The machine learning model was trained using 410 samples collected under various lighting situations, followed by labeling and validation. On 100 samples, it was examined, and the estimated variance was determined to be between 2–11%. Additionally, we used unidentified soil samples to confirm our method, and we estimated an average variance of 8.9–11.7% for the stated nutrients, which is similar to the PUSA STFR kit data.

Enhancing multilayer perceptron neural network using archive-based harris hawks optimizer to predict gold prices

The success of the Multi-Layer Perceptron Neural Network (MLP) relies on carefully configuring its weights and biases to promising values. The gradient descent technique is usually used to select the optimal MLP configuration. Such a technique can stuck in local optima and its convergence towards promising regions in the search space is slow. Harris Hawks optimizer (HHO) is a recently proposed metaheuristic technique that mimics the harris’ hawks behavior to chase and catch prey. In this paper, the algorithm search process is enhanced by introducing an external archive that saves the best solutions to be used in the next iterations. Gold price prediction is essential for future investment planning and decision-making in mining projects. The gold price is predicted using a multilayer perceptron (MLP) neural network (NN) that has the optimal control parameters obtained using the improved version of Harris Hawks optimizer (HHO) with an external archive which is called (AHHO-NN). The input features with a complicated relationship can influence gold price prediction models such as inflation rate, precious metals prices, and other features. In this sense, the literature suggests a specific reduced number of features for the MLP-NN model to yield high-quality prediction results. In addition to the prediction model, two feature reduction methods are used to select the proper list of the input features for the MLP-NN model to forecast gold price namely, Pearson’s correlation and a newly proposed categorized correlation. Surprisingly, new features not mentioned in the literature are discovered and some are discarded. The time series dataset used has been extracted from several sources and pre-processed to fit the proposed model. Furthermore, the prediction results have been evaluated using several measures such as prediction accuracy, Mean square error (MSE), Root mean square error (RMSE), Mean absolute error (MAE), and average absolute relative deviation (AARD). The performance of the proposed AHHO-NN model is compared against four swarm intelligence algorithms (i.e., HHO-NN, JAYA-NN, MFO-NN, PSO-NN) as well as four classical machine learning techniques (i.e., linear regression (LR), MLP Regressor (MLP), RANSAC Regressor (RANSAC), and TheilSen Regressor (TR)). Interestingly, the proposed AHHO-NN is able to comparatively produce high-quality prediction results for gold prices with relatively high accuracy. This proves the viability of the proposed AHHO-NN model as well as the proper selection of the primitive input features.

Data-driven methods for stress field predictions in random heterogeneous materials

Predicting full-field stress responses is of fundamental importance to assessing materials failure and has various engineering applications in design optimization, manufacturing process control, and structural health monitoring. This article develops and evaluates different data-driven methods for efficient and accurate predictions of full stress fields in random heterogeneous materials. The first approach integrates model order reduction of proper orthogonal decomposition (POD) with classical machine learning techniques (K-nearest neighbors, random forest, and artificial neural networks) to predict full-field responses based on POD-reduced coefficients. However, this strategy shows limitations in predicting full stress fields, especially for heterogeneous material inclusions of small size or being close to the domain boundary. After that, two computer vision-based deep learning approaches were developed for full-field predictions. The first one uses a Resnet-based Convolutional Neural Network (CNN), and the second is based on a modified conditional Generative Adversarial Network (cGAN). Two representative example problems were studied: a random heterogeneous material inclusion or a void varying in size and location. In contrast to POD-based classical machine learning, almost invisible differences were found between the entire stress fields in finite element simulations and computer vision-based deep learning (CNN/cGAN) predictions, with significantly reduced mean squared error (MSE) and correlation values (R2) mostly above 0.99. On the other hand, the proposed cGAN provides more accurate predictions than CNN with fewer epochs.

Metaverse and Healthcare: Machine Learning-Enabled Digital Twins of Cancer.

Medical digital twins, which represent medical assets, play a crucial role in connecting the physical world to the metaverse, enabling patients to access virtual medical services and experience immersive interactions with the real world. One serious disease that can be diagnosed and treated using this technology is cancer. However, the digitalization of such diseases for use in the metaverse is a highly complex process. To address this, this study aims to use machine learning (ML) techniques to create real-time and reliable digital twins of cancer for diagnostic and therapeutic purposes. The study focuses on four classical ML techniques that are simple and fast for medical specialists without extensive Artificial Intelligence (AI) knowledge, and meet the requirements of the Internet of Medical Things (IoMT) in terms of latency and cost. The case study focuses on breast cancer (BC), the second most prevalent form of cancer worldwide. The study also presents a comprehensive conceptual framework to illustrate the process of creating digital twins of cancer, and demonstrates the feasibility and reliability of these digital twins in monitoring, diagnosing, and predicting medical parameters.

Pose detection using deep learning for gait assessment

This approach of gait analysis helps in solving the problems associated with the health of aged people by concentrating on the frailty and senility syndromes that affect elderly individuals. Deterioration of cognitive and motor abilities is a common result of ageing and has an effect on elderly people's quality of life. Some studies have connected these changes in gait patterns to the decline in cognitive and motor function. As a result, gait analysis is a useful tool for diagnosing senility and frailty diseases. By transferring the pose estimation data to a correct statistical analysis, gait analysis can be carried out utilizing machine learning and computer vision approach, allowing physiotherapists to analyse and treat patients accordingly. The main problem observed is that physiotherapists will analyse stroke patients' walking patterns on their own. As a result, this approach solves the problem by providing an accurate result of the observation. As everything is computerised, proper treatment can be provided with accurate data. A physiotherapist could also use machine learning classification techniques to analyse and classify the observed disorder and correct it. Our study mainly focuses on using machine learning approach instead of traditional sensor based or consultation just by vision. This will serve as a newer and better proposition of the same.

Deep fiber clustering: Anatomically informed fiber clustering with self-supervised deep learning for fast and effective tractography parcellation

White matter fiber clustering is an important strategy for white matter parcellation, which enables quantitative analysis of brain connections in health and disease. In combination with expert neuroanatomical labeling, data-driven white matter fiber clustering is a powerful tool for creating atlases that can model white matter anatomy across individuals. While widely used fiber clustering approaches have shown good performance using classical unsupervised machine learning techniques, recent advances in deep learning reveal a promising direction toward fast and effective fiber clustering. In this work, we propose a novel deep learning framework for white matter fiber clustering, Deep Fiber Clustering (DFC), which solves the unsupervised clustering problem as a self-supervised learning task with a domain-specific pretext task to predict pairwise fiber distances. This process learns a high-dimensional embedding feature representation for each fiber, regardless of the order of fiber points reconstructed during tractography. We design a novel network architecture that represents input fibers as point clouds and allows the incorporation of additional sources of input information from gray matter parcellation. Thus, DFC makes use of combined information about white matter fiber geometry and gray matter anatomy to improve the anatomical coherence of fiber clusters. In addition, DFC conducts outlier removal naturally by rejecting fibers with low cluster assignment probability. We evaluate DFC on three independently acquired cohorts, including data from 220 individuals across genders, ages (young and elderly adults), and different health conditions (healthy control and multiple neuropsychiatric disorders). We compare DFC to several state-of-the-art white matter fiber clustering algorithms. Experimental results demonstrate superior performance of DFC in terms of cluster compactness, generalization ability, anatomical coherence, and computational efficiency.

A deep convolutional neural network for burn progression mapping using Sentinel-1 SAR time-series

ABSTRACT Forest fires burn various natural ecosystems worldwide and can harm the environment and human life. Accordingly, real-time monitoring of this phenomenon and early decision-making warning is vital. Active remote sensing systems, such as Synthetic Aperture Radar (SAR) sensors, provide an excellent opportunity for burn mapping because they can penetrate through clouds and smoke day and night. In this study, the potential of Sentinel-1 SAR data was investigated by deploying a deep convolutional neural network (CNN) based framework to map burn progression dynamically, in both supervised and transfer manners. Accordingly, an optimized deep architecture was designated to use SAR data features with high sensitivity to map the burned areas. The proposed method includes three main steps: 1) extraction of SAR indices, 2) training deep CNN model with a limited number of scenes and 3) assessing the transferability of the CNN for estimating burn progression for any unseen scene. Sentinel-1 SAR indices of log-ratio, radar burn difference (RBD), and difference of dual-polarization SAR vegetation index (ΔDPSVI) were obtained to be fed to the CNN. To validate the efficiency of the proposed approach, two fire events, i.e. the Derazno fire in Iran (2021) and the Rossomanno-Grottascura-Bellia fire in Italy (2017), were considered. For the scenes including training samples, the proposed method improved the overall accuracies (OAs) of classical machine learning techniques (i.e. SVM and RF) significantly (more than 4%). However, the improvement was minor when compared to a CNN using only log-ratio as the input channel (log-ratio CNN). For the scenes without training samples (unseen dates), the investigated transferred model performed substantially better (3% higher OA) compared to the other machine learning methods and the log-ratio CNN. This finding approves that the obtained SAR indices boost the transferability of the CNN model for burn progression mapping at unseen scenes.

Machine Learning-Based Classification of Asbestos-Containing Roofs Using Airborne RGB and Thermal Imagery

Detecting asbestos-containing roofs has been of great interest in the past few years as the substance negatively affects human health and the environment. Different remote sensing data have been successfully used for this purpose. However, RGB and thermal data have yet to be investigated. This study aims to investigate the classification of asbestos-containing roofs using RGB and airborne thermal data and state-of-the-art machine learning (ML) classification techniques. With the rapid development of ML reflected in this study, we evaluate three classifiers: Random Forest (RF), Support Vector Machine (SVM), and eXtreme Gradient Boosting (XGBoost). We have used several image enhancement techniques to produce additional bands to improve the classification results. For feature selection, we used the Boruta technique; based on the results, we have constructed four different variations of the dataset. The results showed that the most important features for asbestos-containing roof detection were the investigated spectral indices in this study. From a ML point of view, SVM outperformed RF and XGBoost in the dataset using only the spectral indices, with a balanced accuracy of 0.93. Our results showed that RGB bands could produce as accurate results as the multispectral and hyperspectral data with the addition of spectral indices.

Large-scale quantification of groundwater recharge threshold conditions using machine learning classifications: An attempt over the Australian continent

The large-scale quantification of groundwater recharge threshold conditions is a bottleneck in the field of research, considering that groundwater recharge processes can be quite complex and affected by different factors in different regions. Traditional statistical approaches have either focused on local to regional scales or are limited to a single factor such as rainfall, resulting in a lack of a comprehensive understanding of groundwater recharge mechanisms on large scales. Considering that finding a threshold to instantly determine whether the groundwater is in the state of recharge or discharge is a classification task involving multiple factors, machine learning techniques with self-learning capabilities, efficient processing of big data, and mining relationships between multiple factors is, thus, applied to this study. In this study, three traditional machine learning classification techniques are explored; Classification and Regression Tree (CART), Random Forest (RF), and Logistical Regression (LR), to quantify specific threshold values for each condition (including rainfall, evaporation, soil moisture, runoff and vegetation) since they are more interpretable compared to advanced machine learning techniques such as neural networks. Using the Australian continent as an example, CART is the fastest method and provides an average classification accuracy (76%) at almost the same levels compared to RF (77–78%) and LR (79–80%). Meanwhile, CART is the only method to rank the importance of conditions while providing a specific threshold for each condition. When inferring the groundwater recharge mechanism through threshold conditions, the CART method suggests that a primary (the most important) threshold condition is sufficient for most parts of the Australian continent, e.g., rainfall as the primary threshold condition in northern Australia indicates a direct recharge mechanism. Only the coastal areas of southeastern and southwestern Australia are dominated by multiple threshold conditions, showing the complex groundwater recharge mechanism. Overall, the significance of this work is to provide the spatial maps of threshold conditions that mainly control groundwater recharge and their specific threshold values, which are crucial for managers to carry out a good plan over a large area. More importantly, this study demonstrates the potential of traditional machine learning methods to reveal groundwater recharge mechanisms, which may help further refine the development of groundwater models in the future.

Pattern Recognition in Breast Cancer Using Machine Learning

Breast Cancer is the most often identified cancer among women and major reason for increasing mortality rate among women. As the diagnosis of this disease manually takes long hours and the lesser availability of systems, there is a need to develop the automatic diagnosis system for early detection of cancer. Data mining techniques contribute a lot in the development of such system. For the classification of benign and malignant tumor we have used classification techniques of machine learning in which the machine is learned from the past data and can predict the category of new input. This paper is a relative study on the implementation of models using Logistic Regression, Support Vector Machine (SVM) and Latent Dirichlet Allocation (LDA) is done on the dataset taken from the UCI repository. With respect to the results of accuracy, precision, sensitivity, specificity and False Positive Rate the efficiency of each algorithm is measured and compared. Breast Cancer is the most leading malignancy affecting 2.1 million women each year which leads to greatest number of deaths among women. Early treatment not only helps to cure cancer but also helps in prevention of its recurrence. And hence this system mainly focuses on prediction of breast cancer where it uses different machine learning algorithms for creating models like decision tree, logistic regression, random forest which are applied on pre-processed data which suspects greater accuracy for prediction. Amongst all the models, Random Forest Classification leads to best accuracy with 98.6% techniques

Deep Learning based techniques for Neuro-degenerative disorders detection

Mental disorders are neural issues that influence brain cognition and social connectivity. The significant increase in mental disorders needs prompt detection for effective treatment. Psychiatric disorders can be partially diagnosed with constant counseling by using anti-depressants and other inhibitors. However, this is a laborious and also time-consuming process. This leads to investigating the computational techniques that can be applied to detect psychiatric illness in its premature phases. Various works are proposed to detect mental disorders from four data forms: neuro-images, EEG signals, textual data, and gene data. The researchers aim to investigate various models that can detect mental illness using advanced deep learning techniques. Through exploring authoritative databases, the authors gathered papers and studied classic machine learning and primary deep learning techniques for anticipating mental health issues. As the results of various models, the study also presented the difficulties and constraints that machine learning researchers have encountered while studying psychological disorders. The collected articles are categorized into four forms based on the data type used for the analysis. This paper proposes four different hybrid models to detect mental disorders in the early stages from four different data forms. Standardized metrics like area under the receiver operating characteristic curve (AUC–ROC) score and average classification accuracy (ACA) are used to evaluate the models. The proposed models outperformed all the state-of-the-art models in disorder prediction. Furthermore, the study discusses specific suggestions for future studies and development in the use of learning algorithms in the field of mental health.

Condition based Ensemble Deep Learning and Machine Learning Classification Technique for Integrated Potential Fishing Zone Future Forecasting

Artificial Intelligence (AI) technologies have become a popular application in order to improve the sustainability of smart fisheries. Although the ultimate objective of AI applications is often described as sustainability, there is yet no proof as to how AI contributes to sustainable fisheries. The proper monitoring of the longitudinal delivery of different human impacts on activities such as fishing is a major concern today in aquatic conservation. The term "potential fishing zone" (PFZ) refers to an anticipated area of any given sea where a variety of fish may congregate for some time. The forecast is made based on factors including the sea surface temperature (SST) and the sea superficial chlorophyll attentiveness. Fishing advisories are a by-product of the identification procedure. Normalization and preliminary processing are applied to these unprocessed data. The gathered attributes, together with financial derivatives and geometric features, are then utilised to make projections about IPFZ's Technique are used to get the final determination (CECT). In this study, we offer a technique for identifying and mapping fishing activity. Experimentations are performed to validate the efficacy of the CECT method in comparison to machine learning (ML) and deep learning (DL) methods across a variety of measurable parameters. Results showed that CECT obtained 94% accuracy, while Convolutional neural network only managed 92% accuracy on 80% training data and 20% testing data.

Performing Automatic Identification and Staging of Urothelial Carcinoma in Bladder Cancer Patients Using a Hybrid Deep-Machine Learning Approach

Accurate clinical staging of bladder cancer aids in optimizing the process of clinical decision-making, thereby tailoring the effective treatment and management of patients. While several radiomics approaches have been developed to facilitate the process of clinical diagnosis and staging of bladder cancer using grayscale computed tomography (CT) scans, the performances of these models have been low, with little validation and no clear consensus on specific imaging signatures. We propose a hybrid framework comprising pre-trained deep neural networks for feature extraction, in combination with statistical machine learning techniques for classification, which is capable of performing the following classification tasks: (1) bladder cancer tissue vs. normal tissue, (2) muscle-invasive bladder cancer (MIBC) vs. non-muscle-invasive bladder cancer (NMIBC), and (3) post-treatment changes (PTC) vs. MIBC.

Unbinned multivariate observables for global SMEFT analyses from machine learning

Theoretical interpretations of particle physics data, such as the determination of the Wilson coefficients of the Standard Model Effective Field Theory (SMEFT), often involve the inference of multiple parameters from a global dataset. Optimizing such interpretations requires the identification of observables that exhibit the highest possible sensitivity to the underlying theory parameters. In this work we develop a flexible open source frame-work, ML4EFT, enabling the integration of unbinned multivariate observables into global SMEFT fits. As compared to traditional measurements, such observables enhance the sensitivity to the theory parameters by preventing the information loss incurred when binning in a subset of final-state kinematic variables. Our strategy combines machine learning regression and classification techniques to parameterize high-dimensional likelihood ratios, using the Monte Carlo replica method to estimate and propagate methodological uncertainties. As a proof of concept we construct unbinned multivariate observables for top-quark pair and Higgs+Z production at the LHC, demonstrate their impact on the SMEFT parameter space as compared to binned measurements, and study the improved constraints associated to multivariate inputs. Since the number of neural networks to be trained scales quadratically with the number of parameters and can be fully parallelized, the ML4EFT framework is well-suited to construct unbinned multivariate observables which depend on up to tens of EFT coefficients, as required in global fits.