Search results for: feature model

Authors: Peyman Sheikholharam Mashhadi

Abstract:

Difficulties in analyzing real world images in classical image processing and machine vision framework have motivated researchers towards considering the biology-based vision. It is a common belief that mammalian visual cortex has been adapted to the statistics of the real world images through the evolution process. There are two well-known successful models of mammalian visual cortical cells: Independent Component Analysis (ICA) and Independent Subspace Analysis (ISA). In this paper, we statistically analyze the dependencies which remain in the components after applying these models to the natural images. Also, we investigate the response of feature detectors to gratings with various parameters in order to find optimal parameters of the feature detectors. Finally, the selectiveness of feature detectors to phase, in both models is considered.

Keywords: statistics, independent component analysis, independent subspace analysis, phase, natural images

Procedia PDF Downloads 318

Authors: Awais Ahmad Khan, Emad Abouel Nasr, H. M. A. Hussein, Abdulrahman Al-Ahmari

Abstract:

Sheet-metal parts have been widely applied in electronics, communication and mechanical industries in recent decades; but the advancement in sheet-metal part design and manufacturing is still behind in comparison with the increasing importance of sheet-metal parts in modern industry. This paper presents a methodology for automatic extraction of some common 2D internal sheet metal features. The features used in this study are taken from Unipunch ™ catalogue. The extraction process starts with the data extraction from STEP file using an object oriented approach and with the application of suitable algorithms and rules, all features contained in the catalogue are automatically extracted. Since the extracted features include geometry and engineering information, they will be effective for downstream application such as feature rebuilding and process planning.

Keywords: feature extraction, internal features, punch shapes, sheet metal

Procedia PDF Downloads 587

Authors: Kathia Melbouci, Sylvie Naudet Collette, Vincent Gay-Bellile, Omar Ait-Aider, Michel Dhome

Abstract:

In this paper, we handle the problem of real time localization and mapping in indoor environment assisted by a partial prior 3D model, using an RGBD sensor. The proposed solution relies on a feature-based RGBD SLAM algorithm to localize the camera and update the 3D map of the scene. To improve the accuracy and the robustness of the localization, we propose to combine in a local bundle adjustment process, geometric information provided by a prior coarse 3D model of the scene (e.g. generated from the 2D floor plan of the building) along with RGBD data from a Kinect camera. The proposed approach is evaluated on a public benchmark dataset as well as on real scene acquired by a Kinect sensor.

Keywords: SLAM, global localization, 3D sensor, bundle adjustment, 3D model

Procedia PDF Downloads 376

Authors: Krishang Surapaneni

Abstract:

The goal of this project is to determine how to predict important aspects of options, including the ask price. We want to compare different machine learning models to learn the best model and the best hyperparameters for that model for this purpose and data set. Option pricing is a relatively new field, and it can be very complicated and intimidating, especially to inexperienced people, so we want to create a machine learning model that can predict important aspects of an option stock, which can aid in future research. We tested multiple different models and experimented with hyperparameter tuning, trying to find some of the best parameters for a machine-learning model. We tested three different models: a Random Forest Regressor, a linear regressor, and an MLP (multi-layer perceptron) regressor. The most important feature in this experiment is the ask price; this is what we were trying to predict. In the field of stock pricing prediction, there is a large potential for error, so we are unable to determine the accuracy of the models based on if they predict the pricing perfectly. Due to this factor, we determined the accuracy of the model by finding the average percentage difference between the predicted and actual values. We tested the accuracy of the machine learning models by comparing the actual results in the testing data and the predictions made by the models. The linear regression model performed worst, with an average percentage error of 17.46%. The MLP regressor had an average percentage error of 11.45%, and the random forest regressor had an average percentage error of 7.42%

Keywords: finance, linear regression model, machine learning model, neural network, stock price

Procedia PDF Downloads 54

Authors: Kasra Majbouri Yazdi, Adel Majbouri Yazdi, Saeid Khodayi, Jingyu Hou, Wanlei Zhou, Saeed Saedy

Abstract:

Fake news and false information are big challenges of all types of media, especially social media. There is a lot of false information, fake likes, views and duplicated accounts as big social networks such as Facebook and Twitter admitted. Most information appearing on social media is doubtful and in some cases misleading. They need to be detected as soon as possible to avoid a negative impact on society. The dimensions of the fake news datasets are growing rapidly, so to obtain a better result of detecting false information with less computation time and complexity, the dimensions need to be reduced. One of the best techniques of reducing data size is using feature selection method. The aim of this technique is to choose a feature subset from the original set to improve the classification performance. In this paper, a feature selection method is proposed with the integration of K-means clustering and Support Vector Machine (SVM) approaches which work in four steps. First, the similarities between all features are calculated. Then, features are divided into several clusters. Next, the final feature set is selected from all clusters, and finally, fake news is classified based on the final feature subset using the SVM method. The proposed method was evaluated by comparing its performance with other state-of-the-art methods on several specific benchmark datasets and the outcome showed a better classification of false information for our work. The detection performance was improved in two aspects. On the one hand, the detection runtime process decreased, and on the other hand, the classification accuracy increased because of the elimination of redundant features and the reduction of datasets dimensions.

Keywords: clustering, fake news detection, feature selection, machine learning, social media, support vector machine

Procedia PDF Downloads 145

Authors: Hassan Masoumi, Ahad Salimi, Nazanin Barhemmat, Babak Gholami

Abstract:

Artificial neural networks have gained a lot of interest as empirical models for their powerful representational capacity, multi input and output mapping characteristics. In fact, most feed-forward networks with nonlinear nodal functions have been proved to be universal approximates. In this paper, we propose a new supervised method for color image classification based on self organizing feature maps (SOFM). This algorithm is based on competitive learning. The method partitions the input space using self-organizing feature maps to introduce the concept of local neighborhoods. Our image classification system entered into RGB image. Experiments with simulated data showed that separability of classes increased when increasing training time. In additional, the result shows proposed algorithms are effective for color image classification.

Keywords: classification, SOFM algorithm, neural network, neighborhood, RGB image

Procedia PDF Downloads 445

Authors: Yufeng Li, Yinan Ma, Jing Wu, Chengnian Long

Abstract:

Infrared object detection aims at separating small and dim targets from cluttered backgrounds, and its capabilities extend beyond the limits of visible light, making it invaluable in a wide range of applications, such as improving safety, security, efficiency, and functionality. However, existing methods are usually sensitive to the noise of the input infrared image, leading to a decrease in target detection accuracy and an increase in the false alarm rate in high-noise environments. To address this issue, an infrared small target detection algorithm called YOLO-IR is proposed in this paper to improve the robustness to high infrared noise. To address the problem that high noise significantly reduces the clarity and reliability of target features in infrared images, we design a soft-threshold coordinate attention mechanism to improve the model’s ability to extract target features and its robustness to noise. Since the noise may overwhelm the local details of the target, resulting in the loss of small target features during depth down-sampling, we propose a deep and shallow feature fusion neck to improve the detection accuracy. In addition, because the generalized Intersection over Union (IoU)-based loss functions may be sensitive to noise and lead to unstable training in high-noise environments, we introduce a Wasserstein-distance based loss function to improve the training of the model. The experimental results show that YOLO-IR achieves a 5.0% improvement in recall and a 6.6% improvement in the F1 score over the existing state-of-the-art model.

Keywords: infrared small target detection, high noise, robustness, soft-threshold coordinate attention, feature fusion

Procedia PDF Downloads 27

Authors: Neerav Kumar, Naumaan Nayyar, Sharath Kashyap

Abstract:

Most companies see an increase in the adoption of machine learning (ML) applications across internal and external-facing use cases. ML applications vend output either in batch or real-time patterns. A complete batch ML pipeline architecture comprises data sourcing, feature engineering, model training, model deployment, model output vending into a data store for downstream application. Due to unclear role expectations, we have observed that scientists specializing in building and optimizing models are investing significant efforts into building the other components of the architecture, which we do not believe is the best use of scientists’ bandwidth. We propose a system architecture created using AWS services that bring industry best practices to managing the workflow and simplifies the process of model deployment and end-to-end data integration for an ML application. This narrows down the scope of scientists’ work to model building and refinement while specialized data engineers take over the deployment, pipeline orchestration, data quality, data permission system, etc. The pipeline infrastructure is built and deployed as code (using terraform, cdk, cloudformation, etc.) which makes it easy to replicate and/or extend the architecture to other models that are used in an organization.

Keywords: data pipeline, machine learning, AWS, architecture, batch machine learning

Procedia PDF Downloads 31

Authors: Turuganti Venkateswarlu, Jagadeesh Anmala

Abstract:

Excessive runoffs from various non-point source land uses, and other point sources are rapidly contaminating the water quality of streams in the Upper Green River watershed, Kentucky, USA. It is essential to maintain the stream water quality as the river basin is one of the major freshwater sources in this province. It is also important to understand the water quality parameters (WQPs) quantitatively and qualitatively along with their important features as stream water is sensitive to climatic events and land-use practices. In this paper, a model was developed for predicting one of the significant WQPs, Fecal Coliform (FC) from precipitation, temperature, urban land use factor (ULUF), agricultural land use factor (ALUF), and forest land-use factor (FLUF) using Random Forest (RF) algorithm. The RF model, a novel ensemble learning algorithm, can even find out advanced feature importance characteristics from the given model inputs for different combinations. This model’s outcomes showed a good correlation between FC and climate events and land use factors (R2 = 0.94) and precipitation and temperature are the primary influencing factors for FC.

Keywords: water quality, land use factors, random forest, fecal coliform

Procedia PDF Downloads 167

Authors: E. Benga, T. Tengen, A. Alugongo

Abstract:

Periodic and continuous inventory models are the two classical management tools used to handle inventories. These models have advantages and disadvantages. The implementation of both continuous (r,Q) inventory and periodic (R, S) inventory models in most manufacturing plants comes with higher cost. Such high inventory costs are due to the fact that most manufacturing plants are not flexible enough. Since demand and lead-time are two important variables of every inventory models, their effect on the flexibility of the manufacturing plant matter most. Unfortunately, these effects are not clearly understood by managers. The reason is that the decision parameters of the continuous (r, Q) inventory and periodic (R, S) inventory models are not designed to effectively deal with the issues of uncertainties such as poor manufacturing performances, delivery performance supplies performances. There is, therefore, a need to come up with a predictive and hybrid inventory model that can combine in some sense the feature of the aforementioned inventory models. A linear combination technique is used to hybridize both continuous (r, Q) inventory and periodic (R, S) inventory models. The behavior of such hybrid inventory model is described by a differential equation and then optimized. From the results obtained after simulation, the continuous (r, Q) inventory model is more effective than the periodic (R, S) inventory models in the short run, but this difference changes as time goes by. Because the hybrid inventory model is more cost effective than the continuous (r,Q) inventory and periodic (R, S) inventory models in long run, it should be implemented for strategic decisions.

Keywords: periodic inventory, continuous inventory, hybrid inventory, optimization, manufacturing plant

Procedia PDF Downloads 358

Authors: Mehdi Brahimi, Kamal Medjaher, Noureddine Zerhouni, Mohammed Leouatni

Abstract:

The Prognostics and Health Management is a system engineering discipline which provides solutions and models to the implantation of a predictive maintenance. The approach is based on extracting useful information from monitoring data to assess the “health” state of an industrial equipment or an asset. In this paper, we examine multiple extracted features from Pantograph-Catenary contact force in order to select the most relevant ones to achieve a diagnostics function. The feature extraction methodology is based on simulation data generated thanks to a Pantograph-Catenary simulation software called INPAC and measurement data. The feature extraction method is based on both statistical and signal processing analyses. The feature selection method is based on statistical criteria.

Keywords: catenary/pantograph interaction, diagnostics, Prognostics and Health Management (PHM), quality of current collection

Procedia PDF Downloads 260

Authors: Wanhyun Cho, Soonja Kang, Sangkyoon Kim, Soonyoung Park

Abstract:

In this paper, we present the human action recognition method using the variational Bayesian HMM with the Dirichlet process mixture (DPM) of the Gaussian-Wishart emission model (GWEM). First, we define the Bayesian HMM based on the Dirichlet process, which allows an infinite number of Gaussian-Wishart components to support continuous emission observations. Second, we have considered an efficient variational Bayesian inference method that can be applied to drive the posterior distribution of hidden variables and model parameters for the proposed model based on training data. And then we have derived the predictive distribution that may be used to classify new action. Third, the paper proposes a process of extracting appropriate spatial-temporal feature vectors that can be used to recognize a wide range of human behaviors from input video image. Finally, we have conducted experiments that can evaluate the performance of the proposed method. The experimental results show that the method presented is more efficient with human action recognition than existing methods.

Keywords: human action recognition, Bayesian HMM, Dirichlet process mixture model, Gaussian-Wishart emission model, Variational Bayesian inference, prior distribution and approximate posterior distribution, KTH dataset

Procedia PDF Downloads 321

Authors: H. G. C. P. Dinesh, G. Tharshini, M. P. B. Ekanayake, G. M. R. I. Godaliyadda

Abstract:

This paper presents a new spatial feature extraction method based on principle component analysis (PCA) and Fisher Discernment Analysis (FDA) for facial expression recognition. It not only extracts reliable features for classification, but also reduces the feature space dimensions of pattern samples. In this method, first each gray scale image is considered in its entirety as the measurement matrix. Then, principle components (PCs) of row vectors of this matrix and variance of these row vectors along PCs are estimated. Therefore, this method would ensure the preservation of spatial information of the facial image. Afterwards, by incorporating the spectral information of the eigen-filters derived from the PCs, a feature vector was constructed, for a given image. Finally, FDA was used to define a set of basis in a reduced dimension subspace such that the optimal clustering is achieved. The method of FDA defines an inter-class scatter matrix and intra-class scatter matrix to enhance the compactness of each cluster while maximizing the distance between cluster marginal points. In order to matching the test image with the training set, a cosine similarity based Bayesian classification was used. The proposed method was tested on the Cohn-Kanade database and JAFFE database. It was observed that the proposed method which incorporates spatial information to construct an optimal feature space outperforms the standard PCA and FDA based methods.

Keywords: facial expression recognition, principle component analysis (PCA), fisher discernment analysis (FDA), eigen-filter, cosine similarity, bayesian classifier, f-measure

Procedia PDF Downloads 402

Authors: Norah Mohammed Alshahrani, Abdulaziz Almaleh

Abstract:

Autism Spectrum Disorder (ASD) is a condition related to issues with brain development that affects how a person recognises and communicates with others which results in difficulties with interaction and communication socially and it is constantly growing. Early recognition of ASD allows children to lead safe and healthy lives and helps doctors with accurate diagnoses and management of conditions. Therefore, it is crucial to develop a method that will achieve good results and with high accuracy for the measurement of ASD in children. In this paper, ASD datasets of toddlers and children have been analyzed. We employed the following machine learning techniques to attempt to explore ASD and they are Random Forest (RF), Decision Tree (DT), Na¨ıve Bayes (NB) and Support Vector Machine (SVM). Then Feature selection was used to provide fewer attributes from ASD datasets while preserving model performance. As a result, we found that the best result has been provided by the Support Vector Machine (SVM), achieving 0.98% in the toddler dataset and 0.99% in the children dataset.

Keywords: autism spectrum disorder, machine learning, feature selection, support vector machine

Procedia PDF Downloads 113

Authors: Pei-Jyun Hsieh, Cheng-Hsuan Li, Bor-Chen Kuo

Abstract:

For hyperspectral image classification, feature reduction is an important pre-processing for avoiding the Hughes phenomena due to the difficulty for collecting training samples. Hence, lots of researches developed feature selection methods such as F-score, HSIC (Hilbert-Schmidt Independence Criterion), and etc., to improve hyperspectral image classification. However, most of them only consider the class separability in the original space, i.e., a linear class separability. In this study, we proposed a nonlinear class separability measure based on kernel trick for selecting an appropriate feature subset. The proposed nonlinear class separability was formed by a generalized RBF kernel with different bandwidths with respect to different features. Moreover, it considered the within-class separability and the between-class separability. A genetic algorithm was applied to tune these bandwidths such that the smallest with-class separability and the largest between-class separability simultaneously. This indicates the corresponding feature space is more suitable for classification. In addition, the corresponding nonlinear classification boundary can separate classes very well. These optimal bandwidths also show the importance of bands for hyperspectral image classification. The reciprocals of these bandwidths can be viewed as weights of bands. The smaller bandwidth, the larger weight of the band, and the more importance for classification. Hence, the descending order of the reciprocals of the bands gives an order for selecting the appropriate feature subsets. In the experiments, three hyperspectral image data sets, the Indian Pine Site data set, the PAVIA data set, and the Salinas A data set, were used to demonstrate the selected feature subsets by the proposed nonlinear feature selection method are more appropriate for hyperspectral image classification. Only ten percent of samples were randomly selected to form the training dataset. All non-background samples were used to form the testing dataset. The support vector machine was applied to classify these testing samples based on selected feature subsets. According to the experiments on the Indian Pine Site data set with 220 bands, the highest accuracies by applying the proposed method, F-score, and HSIC are 0.8795, 0.8795, and 0.87404, respectively. However, the proposed method selects 158 features. F-score and HSIC select 168 features and 217 features, respectively. Moreover, the classification accuracies increase dramatically only using first few features. The classification accuracies with respect to feature subsets of 10 features, 20 features, 50 features, and 110 features are 0.69587, 0.7348, 0.79217, and 0.84164, respectively. Furthermore, only using half selected features (110 features) of the proposed method, the corresponding classification accuracy (0.84168) is approximate to the highest classification accuracy, 0.8795. For other two hyperspectral image data sets, the PAVIA data set and Salinas A data set, we can obtain the similar results. These results illustrate our proposed method can efficiently find feature subsets to improve hyperspectral image classification. One can apply the proposed method to determine the suitable feature subset first according to specific purposes. Then researchers can only use the corresponding sensors to obtain the hyperspectral image and classify the samples. This can not only improve the classification performance but also reduce the cost for obtaining hyperspectral images.

Keywords: hyperspectral image classification, nonlinear feature selection, kernel trick, support vector machine

Procedia PDF Downloads 241

Authors: Sokkhey Phauk, Takeo Okazaki

Abstract:

The challenging task in educational institutions is to maximize the high performance of students and minimize the failure rate of poor-performing students. An effective method to leverage this task is to know student learning patterns with highly influencing factors and get an early prediction of student learning outcomes at the timely stage for setting up policies for improvement. Educational data mining (EDM) is an emerging disciplinary field of data mining, statistics, and machine learning concerned with extracting useful knowledge and information for the sake of improvement and development in the education environment. The study is of this work is to propose techniques in EDM and integrate it into a web-based system for predicting poor-performing students. A comparative study of prediction models is conducted. Subsequently, high performing models are developed to get higher performance. The hybrid random forest (Hybrid RF) produces the most successful classification. For the context of intervention and improving the learning outcomes, a feature selection method MICHI, which is the combination of mutual information (MI) and chi-square (CHI) algorithms based on the ranked feature scores, is introduced to select a dominant feature set that improves the performance of prediction and uses the obtained dominant set as information for intervention. By using the proposed techniques of EDM, an academic performance prediction system (APPS) is subsequently developed for educational stockholders to get an early prediction of student learning outcomes for timely intervention. Experimental outcomes and evaluation surveys report the effectiveness and usefulness of the developed system. The system is used to help educational stakeholders and related individuals for intervening and improving student performance.

Keywords: academic performance prediction system, educational data mining, dominant factors, feature selection method, prediction model, student performance

Procedia PDF Downloads 85

Authors: Jianhong Xiang, Rui Sun, Linyu Wang

Abstract:

In recent years, due to the continuous improvement of deep learning theory, Convolutional Neural Network (CNN) has played a great superior performance in the research of Hyperspectral Image (HSI) classification. Since HSI has rich spatial-spectral information, only utilizing a single dimensional or single size convolutional kernel will limit the detailed feature information received by CNN, which limits the classification accuracy of HSI. In this paper, we design a multi-scale CNN with MLP based on residual and attention modules (INRAM-3DCNN) for the HSI classification task. We propose to use multiple 3D convolutional kernels to extract the packet feature information and fully learn the spatial-spectral features of HSI while designing residual 3D convolutional branches to avoid the decline of classification accuracy due to network degradation. Secondly, we also design the 2D Inception module with a joint channel attention mechanism to quickly extract key spatial feature information at different scales of HSI and reduce the complexity of the 3D model. Due to the high parallel processing capability and nonlinear global action of the Multilayer Perceptron (MLP), we use it in combination with the previous CNN structure for the final classification process. The experimental results on two HSI datasets show that the proposed INRAM-3DCNN method has superior classification performance and can perform the classification task excellently.

Keywords: INRAM-3DCNN, residual, channel attention, hyperspectral image classification

Procedia PDF Downloads 37

Authors: Xiuying Wang, Changliang Li, Bo Xu

Abstract:

A fixed-length feature vector is required for many machine learning algorithms in NLP field. Word embeddings have been very successful at learning lexical information. However, they cannot capture the compositional meaning of sentences, which prevents them from a deeper understanding of language. In this paper, we introduce a novel hierarchical tree long short-term memory (HTLSTM) model that learns vector representations for sentences of arbitrary syntactic type and length. We propose to split one sentence into three hierarchies: short phrase, long phrase and full sentence level. The HTLSTM model gives our algorithm the potential to fully consider the hierarchical information and long-term dependencies of language. We design the experiments on both English and Chinese corpus to evaluate our model on sentiment analysis task. And the results show that our model outperforms several existing state of the art approaches significantly.

Keywords: deep learning, hierarchical tree long short-term memory, sentence representation, sentiment analysis

Procedia PDF Downloads 328

Authors: Belkacem Soundes, Guezouli Larbi

Abstract:

This paper presents a robust and accurate method for text detection and localization over lecture videos. Frame regions are classified into text or background based on visual feature analysis. However, lecture video shows significant degradation mainly related to acquisition conditions, camera motion and environmental changes resulting in low quality videos. Hence, affecting feature extraction and description efficiency. Moreover, traditional text detection methods cannot be directly applied to lecture videos. Therefore, robust feature extraction methods dedicated to this specific video genre are required for robust and accurate text detection and extraction. Method consists of a three-step process: Slide region detection and segmentation; Feature extraction and non-text filtering. For robust and effective features extraction moment functions are used. Two distinct types of moments are used: orthogonal and non-orthogonal. For orthogonal Zernike Moments, both Pseudo Zernike moments are used, whereas for non-orthogonal ones Hu moments are used. Expressivity and description efficiency are given and discussed. Proposed approach shows that in general, orthogonal moments show high accuracy in comparison to the non-orthogonal one. Pseudo Zernike moments are more effective than Zernike with better computation time.

Keywords: text detection, text localization, lecture videos, pseudo zernike moments

Procedia PDF Downloads 122

Authors: Salim H. Al Salmani, Kevin Tansey, Mohammed S. Ozigis

Abstract:

The study of the growth of built-up areas and settlement expansion is a major exercise that city managers seek to undertake to establish previous and current developmental trends. This is to ensure that there is an equal match of settlement expansion needs to the appropriate levels of services and infrastructure required. This research aims at demonstrating the potential of satellite image processing technique, harnessing the utility of single feature probability-object based image analysis technique in assessing the urban growth dynamics of the Muscat Governorate in Oman for the period 1990, 2002 and 2013. This need is fueled by the continuous expansion of the Muscat Governorate beyond predicted levels of infrastructural provision. Landsat Images of the years 1990, 2002 and 2013 were downloaded and preprocessed to forestall appropriate radiometric and geometric standards. A novel approach of probability filtering of the target feature segment was implemented to derive the spatial extent of the final Built-Up Area of the Muscat governorate for the three years period. This however proved to be a useful technique as high accuracy assessment results of 55%, 70%, and 71% were recorded for the Urban Landcover of 1990, 2002 and 2013 respectively. Furthermore, the Normalized Differential Built – Up Index for the various images were derived and used to consolidate the results of the SFP-OBIA through a linear regression model and visual comparison. The result obtained showed various hotspots where urbanization have sporadically taken place. Specifically, settlement in the districts (Wilayat) of AL-Amarat, Muscat, and Qurayyat experienced tremendous change between 1990 and 2002, while the districts (Wilayat) of AL-Seeb, Bawshar, and Muttrah experienced more sporadic changes between 2002 and 2013.

Keywords: urban growth, single feature probability, object based image analysis, landcover change

Procedia PDF Downloads 250

Authors: Golnush Masghati-Amoli, Paul Chin

Abstract:

Over the past recent years, with the rapid increases in data availability and computing power, Machine Learning (ML) techniques have been called on in a range of different industries for their strong predictive capability. However, the use of Machine Learning in commercial banking has been limited due to a special challenge imposed by numerous regulations that require lenders to be able to explain their analytic models, not only to regulators but often to consumers. In other words, although Machine Leaning techniques enable better prediction with a higher level of accuracy, in comparison with other industries, they are adopted less frequently in commercial banking especially for scoring purposes. This is due to the fact that Machine Learning techniques are often considered as a black box and fail to provide information on why a certain risk score is given to a customer. In order to bridge this gap between the explain-ability and performance of Machine Learning techniques, a Hybrid Model is developed at Dun and Bradstreet that is focused on blending Machine Learning algorithms with traditional approaches such as scorecards. The Hybrid Model maximizes efficiency of traditional scorecards by merging its practical benefits, such as explain-ability and the ability to input domain knowledge, with the deep insights of Machine Learning techniques which can uncover patterns scorecard approaches cannot. First, through development of Machine Learning models, engineered features and latent variables and feature interactions that demonstrate high information value in the prediction of customer risk are identified. Then, these features are employed to introduce observed non-linear relationships between the explanatory and dependent variables into traditional scorecards. Moreover, instead of directly computing the Weight of Evidence (WoE) from good and bad data points, the Hybrid Model tries to match the score distribution generated by a Machine Learning algorithm, which ends up providing an estimate of the WoE for each bin. This capability helps to build powerful scorecards with sparse cases that cannot be achieved with traditional approaches. The proposed Hybrid Model is tested on different portfolios where a significant gap is observed between the performance of traditional scorecards and Machine Learning models. The result of analysis shows that Hybrid Model can improve the performance of traditional scorecards by introducing non-linear relationships between explanatory and target variables from Machine Learning models into traditional scorecards. Also, it is observed that in some scenarios the Hybrid Model can be almost as predictive as the Machine Learning techniques while being as transparent as traditional scorecards. Therefore, it is concluded that, with the use of Hybrid Model, Machine Learning algorithms can be used in the commercial banking industry without being concerned with difficulties in explaining the models for regulatory purposes.

Keywords: machine learning algorithms, scorecard, commercial banking, consumer risk, feature engineering

Procedia PDF Downloads 99

Authors: Abdullah Eqal Al Mazrooei

Abstract:

In this work, a new nonlinear model will be introduced. The model is in the state-space form. The nonlinearity of this model is in the state equation where the state vector is multiplied by its self. This technique makes our model generalizes many famous models as Lotka-Volterra model and Lorenz model which have many applications in the real life. We will apply our new model to estimate the wind speed by using a new nonlinear estimator which suitable to work with our model.

Keywords: nonlinear systems, state-space model, Kronecker product, nonlinear estimator

Procedia PDF Downloads 659

Authors: Nursel Koyuncu, Nihal Ata Tutkun

Abstract:

Frailty model is a survival model that takes into account the unobserved heterogeneity for exploring the relationship between the survival of an individual and several covariates. In the recent years, proposed survival models become more complex and this feature causes convergence problems especially in large data sets. Therefore selection of sample from these big data sets is very important for estimation of parameters. In sampling literature, some authors have defined new sampling schemes to predict the parameters correctly. For this aim, we try to see the effect of sampling design in semi-parametric frailty model. We conducted a simulation study in R programme to estimate the parameters of semi-parametric frailty model for different sample sizes, censoring rates under classical simple random sampling and ranked set sampling schemes. In the simulation study, we used data set recording 17260 male Civil Servants aged 40–64 years with complete 10-year follow-up as population. Time to death from coronary heart disease is treated as a survival-time and age, systolic blood pressure are used as covariates. We select the 1000 samples from population using different sampling schemes and estimate the parameters. From the simulation study, we concluded that ranked set sampling design performs better than simple random sampling for each scenario.

Keywords: frailty model, ranked set sampling, efficiency, simple random sampling

Procedia PDF Downloads 184

Authors: Saahith M. S., Sivakami R.

Abstract:

In the realm of football analytics, particularly focusing on predicting football player performance, the ability to forecast player success accurately is of paramount importance for teams, managers, and fans. This study introduces an elaborate examination of predicting football player performance through the integration of data visualization methods and machine learning algorithms. The research entails the compilation of an extensive dataset comprising player attributes, conducting data preprocessing, feature selection, model selection, and model training to construct predictive models. The analysis within this study will involve delving into feature significance using methodologies like Select Best and Recursive Feature Elimination (RFE) to pinpoint pertinent attributes for predicting player performance. Various machine learning algorithms, including Random Forest, Decision Tree, Linear Regression, Support Vector Regression (SVR), and Artificial Neural Networks (ANN), will be explored to develop predictive models. The evaluation of each model's performance utilizing metrics such as Mean Squared Error (MSE) and R-squared will be executed to gauge their efficacy in predicting player performance. Furthermore, this investigation will encompass a top player analysis to recognize the top-performing players based on the anticipated overall performance scores. Nationality analysis will entail scrutinizing the player distribution based on nationality and investigating potential correlations between nationality and player performance. Positional analysis will concentrate on examining the player distribution across various positions and assessing the average performance of players in each position. Age analysis will evaluate the influence of age on player performance and identify any discernible trends or patterns associated with player age groups. The primary objective is to predict a football player's overall performance accurately based on their individual attributes, leveraging data-driven insights to enrich the comprehension of player success on the field. By amalgamating data visualization and machine learning methodologies, the aim is to furnish valuable tools for teams, managers, and fans to effectively analyze and forecast player performance. This research contributes to the progression of sports analytics by showcasing the potential of machine learning in predicting football player performance and offering actionable insights for diverse stakeholders in the football industry.

Keywords: football analytics, player performance prediction, data visualization, machine learning algorithms, random forest, decision tree, linear regression, support vector regression, artificial neural networks, model evaluation, top player analysis, nationality analysis, positional analysis

Procedia PDF Downloads 15

Authors: Thanakrit Taweesoontorn, Sarucha Yanyong, Poom Konghuayrob

Abstract:

Robotic arms have gained popularity in various industries due to their accuracy and efficiency. This research proposes a method for bin-picking tasks using the Cobot, combining the YOLOv5 CNNs model for object detection and pose estimation with traditional feature detection (FAST), feature description (BRISK), and matching algorithms. By integrating these algorithms and utilizing a small-scale depth sensor camera for capturing depth and color images, the system achieves real-time object detection and accurate pose estimation, enabling the robotic arm to pick objects correctly in both position and orientation. Furthermore, the proposed method is implemented within the ROS framework to provide a seamless platform for robotic control and integration. This integration of robotics, cameras, and AI technology contributes to the development of industrial robotics, opening up new possibilities for automating challenging tasks and improving overall operational efficiency.

Keywords: robotic vision, image processing, applications of robotics, artificial intelligent

Procedia PDF Downloads 49

Authors: Ma Lei-Lei, Zhou You

Abstract:

Aiming at the problems of low detection accuracy of deep learning algorithm in traffic sign detection, this paper proposes improved EfficientDet based traffic sign detection algorithm. Multi-head self-attention is introduced in the minimum resolution layer of the backbone of EfficientDet to achieve effective aggregation of local and global depth information, and this study proposes an improved feature fusion pyramid with increased vertical cross-layer connections, which improves the performance of the model while introducing a small amount of complexity, the Balanced L1 Loss is introduced to replace the original regression loss function Smooth L1 Loss, which solves the problem of balance in the loss function. Experimental results show, the algorithm proposed in this study is suitable for the task of traffic sign detection. Compared with other models, the improved EfficientDet has the best detection accuracy. Although the test speed is not completely dominant, it still meets the real-time requirement.

Keywords: convolutional neural network, transformer, feature pyramid networks, loss function

Procedia PDF Downloads 73

Authors: Ulas Vural, M. Ergun Okay, E. Mesut Yildiz

Abstract:

Telecommunication service providers demand accurate and precise prediction of customer churn probabilities to increase the effectiveness of their customer relation services. The large amount of customer data owned by the service providers is suitable for analysis by machine learning methods. In this study, expenditure data of customers are analyzed by using an artificial neural network (ANN). The ANN model is applied to the data of customers with different billing duration. The proposed model successfully predicts the churn probabilities at 83% accuracy for only three months expenditure data and the prediction accuracy increases up to 89% when the nine month data is used. The experiments also show that the accuracy of ANN model increases on an extended feature set with information of the changes on the bill amounts.

Keywords: customer relationship management, churn prediction, telecom industry, deep learning, artificial neural networks

Procedia PDF Downloads 119

Authors: Adeep Hande, Shubham Agarwal

Abstract:

This paper presents a study on identifying sexism in online texts using various state-of-the-art deep learning models based on BERT. We experimented with different feature sets and model architectures and evaluated their performance using precision, recall, F1 score, and accuracy metrics. We also explored the use of pseudolabeling technique to improve model performance. Our experiments show that the best-performing models were based on BERT, and their multilingual model achieved an F1 score of 0.83. Furthermore, the use of pseudolabeling significantly improved the performance of the BERT-based models, with the best results achieved using the pseudolabeling technique. Our findings suggest that BERT-based models with pseudolabeling hold great promise for identifying sexism in online texts with high accuracy.

Keywords: large language models, semi-supervised learning, sexism detection, data sparsity

Procedia PDF Downloads 37

Authors: Armin Nabaei, M. Omair Ahmad, M. N. S. Swamy

Abstract:

Many facial expression and emotion recognition methods in the traditional approaches of using LDA, PCA, and EBGM have been proposed. In recent years deep learning models have provided a unique platform addressing by automatically extracting the features for the detection of facial expression and emotions. However, deep networks require large training datasets to extract automatic features effectively. In this work, we propose an efficient emotion detection algorithm using face images when only small datasets are available for training. We design a deep network whose feature extraction capability is enhanced by utilizing several parallel modules between the input and output of the network, each focusing on the extraction of different types of coarse features with fined grained details to break the symmetry of produced information. In fact, we leverage long range dependencies, which is one of the main drawback of CNNs. We develop this work by introducing a Dynamic Soft-Margin SoftMax.The conventional SoftMax suffers from reaching to gold labels very soon, which take the model to over-fitting. Because it’s not able to determine adequately discriminant feature vectors for some variant class labels. We reduced the risk of over-fitting by using a dynamic shape of input tensor instead of static in SoftMax layer with specifying a desired Soft- Margin. In fact, it acts as a controller to how hard the model should work to push dissimilar embedding vectors apart. For the proposed Categorical Loss, by the objective of compacting the same class labels and separating different class labels in the normalized log domain.We select penalty for those predictions with high divergence from ground-truth labels.So, we shorten correct feature vectors and enlarge false prediction tensors, it means we assign more weights for those classes with conjunction to each other (namely, “hard labels to learn”). By doing this work, we constrain the model to generate more discriminate feature vectors for variant class labels. Finally, for the proposed optimizer, our focus is on solving weak convergence of Adam optimizer for a non-convex problem. Our noteworthy optimizer is working by an alternative updating gradient procedure with an exponential weighted moving average function for faster convergence and exploiting a weight decay method to help drastically reducing the learning rate near optima to reach the dominant local minimum. We demonstrate the superiority of our proposed work by surpassing the first rank of three widely used Facial Expression Recognition datasets with 93.30% on FER-2013, and 16% improvement compare to the first rank after 10 years, reaching to 90.73% on RAF-DB, and 100% k-fold average accuracy for CK+ dataset, and shown to provide a top performance to that provided by other networks, which require much larger training datasets.

Keywords: computer vision, facial expression recognition, machine learning, algorithms, depp learning, neural networks

Procedia PDF Downloads 51

Authors: Alanoud Moraya Aldalan, Abdulaziz Almaleh

Abstract:

A crucial component of maintaining a customer-oriented business as in the telecom industry is understanding the reasons and factors that lead to customer churn. Competition between telecom companies has greatly increased in recent years. It has become more important to understand customers’ needs in this strong market of telecom industries, especially for those who are looking to turn over their service providers. So, predictive churn is now a mandatory requirement for retaining those customers. Machine learning can be utilized to accomplish this. Churn Prediction has become a very important topic in terms of machine learning classification in the telecommunications industry. Understanding the factors of customer churn and how they behave is very important to building an effective churn prediction model. This paper aims to predict churn and identify factors of customers’ churn based on their past service usage history. Aiming at this objective, the study makes use of feature selection, normalization, and feature engineering. Then, this study compared the performance of four different machine learning algorithms on the Orange dataset: Logistic Regression, Random Forest, Decision Tree, and Gradient Boosting. Evaluation of the performance was conducted by using the F1 score and ROC-AUC. Comparing the results of this study with existing models has proven to produce better results. The results showed the Gradients Boosting with feature selection technique outperformed in this study by achieving a 99% F1-score and 99% AUC, and all other experiments achieved good results as well.

Keywords: machine learning, gradient boosting, logistic regression, churn, random forest, decision tree, ROC, AUC, F1-score

Procedia PDF Downloads 106