Application Of Supervised Learning Research Articles

Fundamental Components and Principles of Supervised Machine Learning Workflows with Numerical and Categorical Data

This paper offers a comprehensive examination of the process involved in developing and automating supervised end-to-end machine learning workflows for forecasting and classification purposes. It offers a complete overview of the components (i.e., feature engineering and model selection), principles (i.e., bias–variance decomposition, model complexity, overfitting, model sensitivity to feature assumptions and scaling, and output interpretability), models (i.e., neural networks and regression models), methods (i.e., cross-validation and data augmentation), metrics (i.e., Mean Squared Error and F1-score) and tools that rule most supervised learning applications with numerical and categorical data, as well as their integration, automation, and deployment. The end goal and contribution of this paper is the education and guidance of the non-AI expert academic community regarding complete and rigorous machine learning workflows and data science practices, from problem scoping to design and state-of-the-art automation tools, including basic principles and reasoning in the choice of methods. The paper delves into the critical stages of supervised machine learning workflow development, many of which are often omitted by researchers, and covers foundational concepts essential for understanding and optimizing a functional machine learning workflow, thereby offering a holistic view of task-specific application development for applied researchers who are non-AI experts. This paper may be of significant value to academic researchers developing and prototyping machine learning workflows for their own research or as customer-tailored solutions for government and industry partners.

Process Parameter Prediction in Laser Powder Bed Fusion Using an Artificial Neural Network

Pores are the inevitable concomitant in the current state of laser powder bed fusion (PBF-LB/M) of AlSI10Mg components. Various pore characteristics, such as pore size and pore shape, influence the quality and affect the intended functionality of the component. Today, the experimental effort to find the appropriate process parameters for additive manufacturing (AM) results in high costs and long time-to-market. Pore formation is highly dependent on the applied process parameters. Consequently, pores can also be seen as an individual process fingerprint. Computed tomography is a commonly used measurement tool for AM components and can be used to comprehensively investigate process-induced defects. Furthermore, X-ray data allows an accurate categorisation of pores and provides a large amount of labelled data for supervised learning applications. The applied classification method classifies the pores into six classes (A-F) according to their shape and size. A total number of 3,066,249 pores detected in cylindrical samples were categorised and used for machine learning. The purpose of this work is to demonstrate an approach for predicting AM process parameters depending on the resulting pore distribution using supervised learning methods. The result is an expandable machine learning model based on an artificial neural network.

Finding Good Attribute Subsets for Improved Decision Trees Using a Genetic Algorithm Wrapper; a Supervised Learning Application in the Food Business Sector for Wine Type Classification

This study aims to provide a method that will assist decision makers in managing large datasets, eliminating the decision risk and highlighting significant subsets of data with certain weight. Thus, binary decision tree (BDT) and genetic algorithm (GA) methods are combined using a wrapping technique. The BDT algorithm is used to classify data in a tree structure, while the GA is used to identify the best attribute combinations from a set of possible combinations, referred to as generations. The study seeks to address the problem of overfitting that may occur when classifying large datasets by reducing the number of attributes used in classification. Using the GA, the number of selected attributes is minimized, reducing the risk of overfitting. The algorithm produces many attribute sets that are classified using the BDT algorithm and are assigned a fitness number based on their accuracy. The fittest set of attributes, or chromosomes, as well as the BDTs, are then selected for further analysis. The training process uses the data of a chemical analysis of wines grown in the same region but derived from three different cultivars. The results demonstrate the effectiveness of this innovative approach in defining certain ingredients and weights of wine’s origin.

On the training sample size and classification performance: An experimental evaluation in seismic facies classification

Machine learning algorithms (MLAs) perform better when enough high-quality training data is provided. However, a lack of training data is frequent in seismic facies classification and many other supervised learning applications. Data labeling for seismic facies classification is time-consuming and requires considerable effort from the domain knowledge expert. This study investigates the effect of training data size on the performance of three popular supervised MLAs used for seismic facies classification. We labeled slices from two seismic datasets of diverse geologic environments and varying classification complexity. AN Field in Malay Basin represents a simple classification problem with three classes, whereas a more complex six classes classification is defined in the Dangerous Grounds (DG) dataset offshore Sabah. The labeled data were constantly reduced by half, resulting in eight training subsets of varying sizes. We trained and evaluated support vector machine (SVM), random forest (RF), and neural network (NN) models using a 10-fold cross-validation (CV) procedure. Performance metrics were computed to study the change in performance in response to the training data size. The experimental results show that, for the DG dataset, where the classification is complex due to the heterogeneous geology and a more number of classes, the larger the training subset, the better the classification performance. Nevertheless, for the simple classification scenario of the AN dataset, the classifiers reached a performance plateau when trained on limited samples. We found that the NN model is the best performer on large datasets. The RF classifier performed well in both datasets. It proved to be robust when trained on limited samples of the DG data. The SVM performed the best where there was a clear margin of separation between the defined classes (the AN data). In contrast, it performed poorly on the DG data and exhibited a performance decline on the AN large subsets.

Information set supported deep learning architectures for improving noisy image classification

Deep learning models have been widely used in many supervised learning applications. However, these models suffer from overfitting due to various types of uncertainty with deteriorating performance when facing data biases, class imbalance, or noise propagation. The Information-Set Deep learning (ISDL) architectures with four variants are developed by integrating information set theory and deep learning principles to address the critical problem of the absence of robust deep learning models. There is a description of the ISDL architectures, learning algorithms, and analytic workflows. The performance of the ISDL models and standard architectures is evaluated using a noise-corrupted benchmark dataset. The experimental results show that the ISDL models can efficiently handle noise-dominated uncertainty and outperform peer architectures.

Tensor Multi-Clustering Parallel Intelligent Computing Method Based on Tensor Chain Decomposition.

Adaptable methods for representing higher-order data with various features and high dimensionality have been demanded by the increasing usage of multi-sensor technologies and the emergence of large data sets. Arrays of multi-dimensional data, known as tensors, can be found in a variety of applications. Standard data that depicts things from a single point of view lacks the semantic richness, utility, and complexity of multi-dimensional data. Research into multi-clustering has taken off since traditional clustering methods are unable to handle large datasets. There are three main kinds of multi-clustering algorithms: Self-weighted Multiview Clustering (SwMC), Latent Multi-view Subspace Clustering (LMSC), and Multi-view Subspace Clustering with Intactness-Aware Similarity (MSC IAS) that are explored in this paper. To evaluate their performance, we do in-depth tests on seven real-world datasets. The three most important metrics Accuracy (ACC), normalized mutual information (NMI), and purity are grouped. Furthermore, traditional Principal Component Analysis (PCA) cannot uncover hidden components within multi-dimensional data. For this purpose, tensor decomposition algorithms have been presented that are flexible in terms of constraint selection and extract more broad latent components. In this examination, we also go through the various tensor decomposition methods, with an emphasis on the issues that classical PCA is designed to solve. Various tensor models are also tested for dimensionality reduction and supervised learning applications in the experiments presented here.

Iterative spatial leave-one-out cross-validation and gap-filling based data augmentation for supervised learning applications in marine remote sensing

ABSTRACT In marine remote sensing, supervised learning can link variables measured in-situ near the ocean surface to variables that can be measured from space. However, the in-situ data used for training and validating such empirical satellite algorithms are often spatially auto-correlated and clustered, giving rise to various statistical challenges such as overfitting to spatial structures. Furthermore, co-located in-situ and satellite measurements are rare in the oceans because of the cost of data collection from research vessels and frequent cloud cover. We propose two methods to mitigate these challenges. The first method builds on spatial leave-one-out cross-validation (SLOOCV), an approach designed to provide sound error estimates when data are spatially auto-correlated by enforcing a minimum separation distance between training and test observations. However, estimating this distance may be impossible with sparse and spatially clustered data. We hence propose to iterate and integrate error estimates over a range of separation distances (iSLOOCV). To address the often-small size of labeled data sets based on marine in-situ data, we tested if increasing the number of observations for algorithm training by means of cloud-filling algorithms for marine satellite data improved predictions. The potential of these two methods is demonstrated by developing empirical algorithms for mapping the proportions of seven diagnostic pigments (DPs) that serve as proxies for phytoplankton community composition in the northern Gulf of Mexico. We estimated the prediction accuracy of 13 algorithms with iSLOOCV, using various sets of satellite data products as input, and found adequate algorithms for 4 of the 7 DPs. Random forests combining ocean color and environmental variables as input had the lowest prediction errors overall. Correlations between predictions and observations estimated by iSLOOCV ranged from 0.69 to 0.85 and mean absolute errors from 0.02 to 0.13. Daily maps and longer-term composites of these DPs were broadly consistent with previously published results. Overall, errors increased when extrapolating over larger distances, highlighting how iSLOOCV can illuminate changes in algorithm performance based on sub-regional data coverage. Generating larger training sets by prior gap-filling substantially improved all error measures for 3 of the 7 DPs, with mixed results for the others. Therefore, data augmentation by gap-filling of satellite data should not be used as a default approach but can be a useful tool when supervised learning applications are suspected to be limited by the size of the training set.

0112 Classification of reconstructed depth profiles shows Global and non-Global slow oscillations differentiate in the hippocampus and thalamus

Abstract Introduction Sleep slow oscillations (SOs, 0.5-1.5 Hz) can be classified on the scalp as Global, Local or Frontal, where Global SOs are found in most electrodes within a short time delay and gate long-range information flow during NREM sleep. In this study, we estimate the current density within the brain that generates a Global SO, to evaluate which sub-cortical structures are involved in Global SO dynamics. We then train multiple machine learning algorithms to distinguish between Global SOs and other SO types, and probe variance of Global/non-Global SO profiles within and across subjects.Sleep slow oscillations (SOs, 0.5-1.5 Hz) can be classified on the scalp as Global, Local or Frontal, where Global SOs are found in most electrodes within a short time delay and gate long-range information flow during NREM sleep. In this study, we estimate the current density within the brain that generates a Global SO, to evaluate which sub-cortical structures are involved in Global SO dynamics. We then train multiple machine learning algorithms to distinguish between Global SOs and other SO types, and probe variance of Global/non-Global SO profiles within and across subjects. Methods 32 volunteers (18 females) slept in the lab with polysomnography including 24 head EEG channels; their sleep was scored according to AASM criteria. SOs were algorithmically detected at each channel and classified as Global or non-Global using our method (Malerba et al., 2019). The depth profile of each SO was reconstructed with current source estimation (in Brainstorm followed by sLORETA), with a standardized head model including 17 regions. Each depth profile was embedded in a matrix averaging current by region and in three 200ms-long time bins: before, during and after the SO trough. Thirty classifiers were applied to this dataset, leveraging Matlab’s supervised learning application. We compared accuracy within and across subjects and identified best-performing algorithms across dataset size. We then used univariate feature selection to quantify the relevance of each region-time pair to successful classification. Results Global/non-Global SOs current depth profiles have higher variance across subjects, and accuracy improves when data is sampled between rather than within individuals. Ensemble subspace methods reached highest accuracy (98.5%). Feature selectivity identified cortical, hippocampal, and thalamic currents at the trough of the SO as the most relevant for Global/non-Global SO classifications. Conclusion We introduce an analytical framework enabling the study of SO depth profiles, including their time evolution, as matrices. The predominant differentiation of Global/non-Global SOs in cortical, hippocampal, and thalamic currents supports the potential functional difference of these SO types. Support (If Any) NIH grant (R01 AG046646) to S.C.M.

Technique Improves Neural Network Training To Quantify Rock Properties From Seismic

This article, written by JPT Technology Editor Chris Carpenter, contains highlights of paper SPE 202685, “Rock-Physics-Guided Data-Science Technique To Improve Neural Network Training To Accurately Quantify Rock Properties From Seismic Data,” by Shraddha Chatterjee, Tanya Colwell, and Jon Downton, CGG, et al. The paper has not been peer reviewed. The success or failure of neural networks in discerning rock properties through a quantitative approach from seismic data and associated seismic attributes is dependent on the amount of available data. Past researchers developed a hybrid theory-guided-data-science (TGDS) technique that augmented the amount of data used for training. This paper shows the application of this method of TGDS-based synthetic seismic catalog generation followed by the application of a deep feed-forward neural network (DFNN) to a real seismic data set for a commercially oil-producing field in the North Sea. Introduction Machine-learning algorithms have been used for several applications in the field of reservoir characterization. Using neural network techniques for supervised learning applications to predict log properties from seismic and seismic attributes is an established technique. However, a known limitation of neural networks is that they depend greatly on the amounts of training data, also called labeled data. Generally, the source of the labeled data during geophysical applications is well-log data. Hence, a need to increase the amount of labeled data was required to increase the accuracy of predictions. The accuracy of neural networks suffers because of the lack of labeled data, especially in exploration study areas where well data are scarce. The hybrid TGDS model augments the amount of data used to train the neural network. During the process, theory-based generated outputs are used as inputs to the data-science component of this work flow. The theory-based components are generated models of rock physics that help in estimating the elastic parameters of a rock caused by changes in fluid and other petrophysical rock properties; this approach then is used to generate a large set of pseudowells. In the next step, synthetic seismic gathers are generated based on the large set of generated pseudowells that are used for training a DFNN. The parameters that are derived for training the neural network are further applied on a real data set. In the complete paper, the authors show an example of this work flow in which it was applied to a North Sea field that produces commercial volumes of oil. The reservoir of interest is of the Paeocene age and is a remobilized injectite sand that cross-cuts the stratigraphy at high angles. The injectite sands can be excellent pay zones with good porosity and permeability, even though their imaging can be challenging.

Asymmetric Error Control Under Imperfect Supervision: A Label-Noise-Adjusted Neyman–Pearson Umbrella Algorithm

Label noise in data has long been an important problem in supervised learning applications as it affects the effectiveness of many widely used classification methods. Recently, important real-world applications, such as medical diagnosis and cybersecurity, have generated renewed interest in the Neyman–Pearson (NP) classification paradigm, which constrains the more severe type of error (e.g., the Type I error) under a preferred level while minimizing the other (e.g., the Type II error). However, there has been little research on the NP paradigm under label noise. It is somewhat surprising that even when common NP classifiers ignore the label noise in the training stage, they are still able to control the Type I error with high probability. However, the price they pay is excessive conservativeness of the Type I error and hence a significant drop in power (i.e., 1 - Type II error). Assuming that domain experts provide lower bounds on the corruption severity, we propose the first theory-backed algorithm that adapts most state-of-the-art classification methods to the training label noise under the NP paradigm. The resulting classifiers not only control the Type I error with high probability under the desired level but also improve power.

Overview of Algorithms for Natural Language Processing and Time Series Analyses.

A host of machine learning algorithms have been used to perform several different tasks in NLP and TSA. Prior to implementing these algorithms, some degree of data preprocessing is required. Deep learning approaches utilizing multilayer perceptrons, recurrent neural networks (RNNs), and convolutional neural networks (CNNs) represent commonly used techniques. In supervised learning applications, all these models map inputs into a predicted output and then model the discrepancy between predicted values and the real output according to a loss function. The parameters of the mapping function are then optimized through the process of gradient descent and backward propagation in order to minimize this loss. This is the main premise behind many supervised learning algorithms. As experience with these algorithms grows, increased applications in the fields of medicine and neuroscience are anticipated.

Intentional deep overfit learning (IDOL): A novel deep learning strategy for adaptive radiation therapy.

Applications of deep learning (DL) are essential to realizing an effective adaptive radiotherapy (ART) workflow. Despite the promise demonstrated by DL approaches in several critical ART tasks, there remain unsolved challenges to achieve satisfactory generalizability of a trained model in a clinical setting. Foremost among these is the difficulty of collecting a task-specific training dataset with high-quality, consistent annotations for supervised learning applications. In this study, we propose a tailored DL framework for patient-specific performance that leverages the behavior of a model intentionally overfitted to a patient-specific training dataset augmented from the prior information available in an ART workflow-an approach we term Intentional Deep Overfit Learning (IDOL). Implementing the IDOL framework in any task in radiotherapy consists of two training stages: (1) training a generalized model with a diverse training dataset of patients, just as in the conventional DL approach, and (2) intentionally overfitting this general model to a small training dataset-specific the patient of interest ( ) generated through perturbations and augmentations of the available task- and patient-specific prior information to establish a personalized IDOL model. The IDOL framework itself is task-agnostic and is, thus, widely applicable to many components of the ART workflow, three of which we use as a proof of concept here: the autocontouring task on replanning CTs for traditional ART, the MRI super-resolution (SR) task for MRI-guided ART, and the synthetic CT (sCT) reconstruction task for MRI-only ART. In the replanning CT autocontouring task, the accuracy measured by the Dice similarity coefficient improves from 0.847 with the general model to 0.935 by adopting the IDOL model. In the case of MRI SR, the mean absolute error (MAE) is improved by 40% using the IDOL framework over the conventional model. Finally, in the sCT reconstruction task, the MAE is reduced from 68 to 22 HU by utilizing the IDOL framework. In this study, we propose a novel IDOL framework for ART and demonstrate its feasibility using three ART tasks. We expect the IDOL framework to be especially useful in creating personally tailored models in situations with limited availability of training data but existing prior information, which is usually true in the medical setting in general and is especially true in ART.

Beyond optimization—supervised learning applications in relativistic laser-plasma experiments

We explore the applications of a variety of machine learning techniques in relativistic laser-plasma experiments beyond optimization purposes. With the trained supervised learning models, the beam charge of electrons produced in a laser wakefield accelerator is predicted given the laser wavefront change caused by a deformable mirror. Feature importance analysis using the trained models shows that specific aberrations in the laser wavefront are favored in generating higher beam charges, which reveals more information than the genetic algorithms and the statistical correlation do. The predictive models enable operations beyond merely searching for an optimal beam charge. The quality of the measured data is characterized, and anomaly detection is demonstrated. The model robustness against measurement errors is examined by applying a range of virtual measurement error bars to the experimental data. This work demonstrates a route to machine learning applications in a highly nonlinear problem of relativistic laser-plasma interaction for in-depth data analysis to assist physics interpretation.

Adaptive local kernels formulation of mutual information with application to active post-seismic building damage inference

The abundance of training data is not guaranteed in various supervised learning applications. One of these situations is the post-earthquake regional damage assessment of buildings. Querying the damage label of each building requires a thorough inspection by experts, and thus, is an expensive task. A practical approach is to sample the most informative buildings in a sequential learning scheme. Active learning methods recommend the most informative cases that are able to maximally reduce the generalization error. The information-theoretic measure of mutual information (MI), which maximizes the expected information gain over the input domain, can be used for informative sampling of a dataset in a pool-based scenario. However, the computational complexity of the standard MI algorithm prevents the utilization of this method on large datasets. A local kernels strategy was proposed to reduce the computational costs, but the adaptability of the kernels to the observed labels was not considered in the original formulation of this strategy. In this article, an adaptive local kernels methodology is developed that enables the conformability of the kernels to the observed output data while enhancing the computational complexity of the standard MI algorithm. The proposed algorithm is developed to work with a Gaussian process regression (GPR) method, where the kernel hyperparameters are updated after each label query using maximum likelihood estimation. In the sequential learning procedure, the updated hyperparameters can be used in the MI kernel matrices to improve the sample suggestion performance. The advantages of the proposed method are demonstrated in a simulation of the 2018 Anchorage, AK, earthquake. It is shown that while the proposed algorithm enables GPR to reach acceptable performance using fewer training data, the computational demand remains lower than the standard local kernels strategy.

AMF: Aggregated Mondrian Forests for Online Learning

AbstractRandom forest (RF) is one of the algorithms of choice in many supervised learning applications, be it classification or regression. The appeal of such tree-ensemble methods comes from a combination of several characteristics: a remarkable accuracy in a variety of tasks, a small number of parameters to tune, robustness with respect to features scaling, a reasonable computational cost for training and prediction, and their suitability in high-dimensional settings. The most commonly used RF variants, however, are ‘offline’ algorithms, which require the availability of the whole dataset at once. In this paper, we introduce AMF, an online RF algorithm based on Mondrian Forests. Using a variant of the context tree weighting algorithm, we show that it is possible to efficiently perform an exact aggregation over all prunings of the trees; in particular, this enables to obtain a truly online parameter-free algorithm which is competitive with the optimal pruning of the Mondrian tree, and thus adaptive to the unknown regularity of the regression function. Numerical experiments show that AMF is competitive with respect to several strong baselines on a large number of datasets for multi-class classification.

TrainSel: An R Package for Selection of Training Populations.

A major barrier to the wider use of supervised learning in emerging applications, such as genomic selection, is the lack of sufficient and representative labeled data to train prediction models. The amount and quality of labeled training data in many applications is usually limited and therefore careful selection of the training examples to be labeled can be useful for improving the accuracies in predictive learning tasks. In this paper, we present an R package, TrainSel, which provides flexible, efficient, and easy-to-use tools that can be used for the selection of training populations (STP). We illustrate its use, performance, and potentials in four different supervised learning applications within and outside of the plant breeding area.

Simulation annealing diagnosis algorithm method for optimized forecast of the dynamic response of floating offshore wind turbines

Design of floating offshore wind turbines (FOWTs) needs reliable and innovative technologies to overcome the challenges on how to better predict the dynamic responses in terms of aero-hydro-servo-elastic disciplines. This paper aims to demonstrate the optimized prediction of the dynamic response of FOWTs by Simulation annealing diagnosis algorithm (SADA). SADA is an Artificial Intelligence technology-based method, which utilizes the advantages of numerical simulation, basin experiment and machine learning algorithms. The actor network in deep deterministic policy gradient (DDPG) is adopted to take actions to adjust the Key disciplinary parameters (KDPs) in each loop according to the feedback of 6DOF motions of platform in dynamic response analysis. The results demonstrated that the mean values of the platform’s motions and rotor axial thrust force could be predicted with higher accuracy. On this basis, other physical quantities that designers are more concerned about but cannot be obtained from experiments and actual measurements will be predicted by SADA with more credibility. This SADA method differs from traditional supervised learning applications in renewable energy, which do not need to be provided physical quantities with strong direct correlation. All targets can be artificially set for SADA to obtain a better self-learning performance. In general, designers can use SADA to get a more accurate and optimized prediction of the dynamic response of FOWTs, especially those physical quantities that cannot be directly obtained through the basin experiments.

Ferroelectric-Like Non-Volatile FET With Amorphous Gate Insulator for Supervised Learning Applications

We experimentally demonstrate ferroelectric-like non-volatile field-effect transistor (NVFET) with the amorphous Al2O3 gate insulator for artificial synapse applications. The switchable polarization (P) is attributed to the voltage modulation of mobile ions in the gate insulator. The ferroelectric-like NVFETs integrated with 3 nm and 6 nm-thick Al2O3 dielectrics demonstrate the capability to mimic various synaptic behaviors including long-term potentiation (LTP), long-term depression (LTD), and spike-timing-dependent plasticity (STDP), under different types of electrical stimuli to the gate electrode. To verify the application of the ferroelectric-like transistors in the Spike Neural Network (SNN), the online training has been carried out based on the synaptic characteristics of the devices, and a decent accuracy (>80%) is achieved for fixed-amplitude ±3 V/100 ns potentiation/depression pulses.

Siamese Neural Network-Based Supervised Slow Feature Extraction for Soft Sensor Application

Slow feature analysis (SFA) is an unsupervised learning method that extracts the latent variables from a time series dataset based on the temporal slowness aspect. Neural networks, owing to their ability to model complex nonlinearities, can be used to extract slow features (SFs) from a dataset obtained from a complicated process. Siamese neural networks can be used for this purpose. Siamese neural networks have a provision of handling two samples at a time and this aspect helps in extracting SFs. For supervised learning applications, the extracted SFs should help predict the outputs. In this article, we present two approaches that extract SFs using Siamese neural networks. The output relevance aspect is brought into feature extraction as a regularization term in the objective function of the Siamese neural network. Such regularization improves the performance of the neural network model. The proposed approaches are implemented on three datasets to demonstrate their effectiveness.

Data Mining Application on Weather Prediction Using Classification Tree, Naïve Bayes and K-Nearest Neighbor Algorithm With Model Testing of Supervised Learning Probabilistic Brier Score, Confusion Matrix and ROC

<p class="AbstractL-MAG">—<em> One of data mining techniques is Classification, used to predict relationships between data on a dataset. The prediction performed by classifying data into several different classes considering certain factor. Classification is a performance of Supervised Learning application where the training data already has a label when entered as input data. Classification is an approach of empirical techniques that can be utilized for short-term weather prediction. The most widely used algorithms in Classification Techniques are Classification Tree, Naïve Bayes and K-Nearest Neighbors. In this study, the author used these three algorithms to predict rain with validation parameters of Brier Score, Confusion Matrix and ROC curves. The input data is synoptic data of Kemayoran Meteorological Station, Jakarta (96745) for 10 years (2006 - 2015) consists of 3528 datasets and 8 attributes. Based on a series of data processing, selection and model testing shows that the Naïve Bayes Algorithm has the best accuracy rate of 77.1% with the category of fair classification so it is quite potential to be used in the operational. The dominant weather attributes in rain formation are moisture (RHavg), minimum temperature (Tmin), maximum temperature (Tmax), average temperature (Tavg) and wind direction (ddd).</em></p>