The aim of this study is to develop a system which can identify and sort peaberries automatically at low cost for coffee producers in developing countries. In this paper, the focus is on the classification of peaberries and normal coffee beans using image processing and machine learning techniques. The peaberry is not bad and not a normal bean. The peaberry is born in an only single seed, relatively round seed from a coffee cherry instead of the usual flat-sided pair of beans. It has another value and flavor. To make the taste of the coffee better, it is necessary to separate the peaberry and normal bean before green coffee beans roasting. Otherwise, the taste of total beans will be mixed, and it will be bad. In roaster procedure time, all the beans shape, size, and weight must be unique; otherwise, the larger bean will take more time for roasting inside. The peaberry has a different size and different shape even though they have the same weight as normal beans. The peaberry roasts slower than other normal beans. Therefore, neither technique provides a good option to select the peaberries. Defect beans, e.g., sour, broken, black, and fade bean, are easy to check and pick up manually by hand. On the other hand, the peaberry pick up is very difficult even for trained specialists because the shape and color of the peaberry are similar to normal beans. In this study, we use image processing and machine learning techniques to discriminate the normal and peaberry bean as a part of the sorting system. As the first step, we applied Deep Convolutional Neural Networks (CNN) and Support Vector Machine (SVM) as machine learning techniques to discriminate the peaberry and normal bean. As a result, better performance was obtained with CNN than with SVM for the discrimination of the peaberry. The trained artificial neural network with high performance CPU and GPU in this work will be simply installed into the inexpensive and low in calculation Raspberry Pi system. We assume that this system will be used in under developed countries. The study evaluates and compares the feasibility of the methods in terms of accuracy of classification and processing speed.
Engagement is one of the most important factors in determining successful outcomes and deep learning in students. Existing approaches to detect student engagement involve periodic human observations that are subject to inter-rater reliability. Our solution uses real-time multimodal multisensor data labeled by objective performance outcomes to infer the engagement of students. The study involves four students with a combined diagnosis of cerebral palsy and a learning disability who took part in a 3-month trial over 59 sessions. Multimodal multisensor data were collected while they participated in a continuous performance test. Eye gaze, electroencephalogram, body pose, and interaction data were used to create a model of student engagement through objective labeling from the continuous performance test outcomes. In order to achieve this, a type of continuous performance test is introduced, the Seek-X type. Nine features were extracted including high-level handpicked compound features. Using leave-one-out cross-validation, a series of different machine learning approaches were evaluated. Overall, the random forest classification approach achieved the best classification results. Using random forest, 93.3% classification for engagement and 42.9% accuracy for disengagement were achieved. We compared these results to outcomes from different models: AdaBoost, decision tree, k-Nearest Neighbor, naïve Bayes, neural network, and support vector machine. We showed that using a multisensor approach achieved higher accuracy than using features from any reduced set of sensors. We found that using high-level handpicked features can improve the classification accuracy in every sensor mode. Our approach is robust to both sensor fallout and occlusions. The single most important sensor feature to the classification of engagement and distraction was shown to be eye gaze. It has been shown that we can accurately predict the level of engagement of students with learning disabilities in a real-time approach that is not subject to inter-rater reliability, human observation or reliant on a single mode of sensor input. This will help teachers design interventions for a heterogeneous group of students, where teachers cannot possibly attend to each of their individual needs. Our approach can be used to identify those with the greatest learning challenges so that all students are supported to reach their full potential.
Generative Adversarial Net (GAN) has proved to be a powerful machine learning tool in image data analysis and generation. In this paper, we propose to use Conditional Generative Adversarial Net (CGAN) to learn and simulate time series data. The conditions include both categorical and continuous variables with different auxiliary information. Our simulation studies show that CGAN has the capability to learn different types of normal and heavy-tailed distributions, as well as dependent structures of different time series. It also has the capability to generate conditional predictive distributions consistent with training data distributions. We also provide an in-depth discussion on the rationale behind GAN and the neural networks as hierarchical splines to establish a clear connection with existing statistical methods of distribution generation. In practice, CGAN has a wide range of applications in market risk and counterparty risk analysis: it can be applied to learn historical data and generate scenarios for the calculation of Value-at-Risk (VaR) and Expected Shortfall (ES), and it can also predict the movement of the market risk factors. We present a real data analysis including a backtesting to demonstrate that CGAN can outperform Historical Simulation (HS), a popular method in market risk analysis to calculate VaR. CGAN can also be applied in economic time series modeling and forecasting. In this regard, we have included an example of hypothetical shock analysis for economic models and the generation of potential CCAR scenarios by CGAN at the end of the paper.
In the field of machine learning, the ensemble has been employed as a common methodology to improve the performance upon multiple base classifiers. However, the true predictions are often canceled out by the false ones during consensus due to a phenomenon called “curse of correlation” which is represented as the strong interferences among the predictions produced by the base classifiers. In addition, the existing practices are still not able to effectively mitigate the problem of imbalanced classification. Based on the analysis on our experiment results, we conclude that the two problems are caused by some inherent deficiencies in the approach of consensus. Therefore, we create an enhanced ensemble algorithm which adopts a designed rank-based chain-mode consensus to overcome the two problems. In order to evaluate the proposed ensemble algorithm, we employ a well-known benchmark data set NSL-KDD (the improved version of dataset KDDCup99 produced by University of New Brunswick) to make comparisons between the proposed and 8 common ensemble algorithms. Particularly, each compared ensemble classifier uses the same 22 base classifiers, so that the differences in terms of the improvements toward the accuracy and reliability upon the base classifiers can be truly revealed. As a result, the proposed rank-based chain-mode consensus is proved to be a more effective ensemble solution than the traditional consensus approach, which outperforms the 8 ensemble algorithms by 20% on almost all compared metrices which include accuracy, precision, recall, F1-score and area under receiver operating characteristic curve.
The inability of organizations to put in place management control measures for Internet of Things (IoT) complexities persists to be a risk concern. Policy makers have been left to scamper in finding measures to combat these security and privacy concerns. IoT forensics is a cumbersome process as there is no standardization of the IoT products, no or limited historical data are stored on the devices. This paper highlights why IoT forensics is a unique adventure and brought out the legal challenges encountered in the investigation process. A quadrant model is presented to study the conflicting aspects in IoT forensics. The model analyses the effectiveness of forensic investigation process versus the admissibility of the evidence integrity; taking into account the user privacy and the providers’ compliance with the laws and regulations. Our analysis concludes that a semi-automated forensic process using machine learning, could eliminate the human factor from the profiling and surveillance processes, and hence resolves the issues of data protection (privacy and confidentiality).
Forecasting the stock market is a very challenging task. Various economic indicators such as GDP, exchange rates, interest rates, and unemployment have a substantial impact on the stock market. Time series models are the traditional methods used to predict stock market changes. In this paper, a machine learning method, Bayesian Additive Regression Tree (BART) is used in predicting stock market indexes based on multiple economic indicators. BART can be used to model heterogeneous treatment effects, and thereby works well when models are misspecified. It also has the capability to handle non-linear main effects and multi-way interactions without much input from financial analysts. In this research, BART is proposed to provide a reliable prediction on day-to-day stock market activities. By comparing the analysis results from BART and with time series method, BART can perform well and has better prediction capability than the traditional methods.
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.
Learning from very big datasets is a significant problem for most present data mining and machine learning algorithms. MicroRNA (miRNA) is one of the important big genomic and non-coding datasets presenting the genome sequences. In this paper, a hybrid method for the classification of the miRNA data is proposed. Due to the variety of cancers and high number of genes, analyzing the miRNA dataset has been a challenging problem for researchers. The number of features corresponding to the number of samples is high and the data suffer from being imbalanced. The feature selection method has been used to select features having more ability to distinguish classes and eliminating obscures features. Afterward, a Convolutional Neural Network (CNN) classifier for classification of cancer types is utilized, which employs a Genetic Algorithm to highlight optimized hyper-parameters of CNN. In order to make the process of classification by CNN faster, Graphics Processing Unit (GPU) is recommended for calculating the mathematic equation in a parallel way. The proposed method is tested on a real-world dataset with 8,129 patients, 29 different types of tumors, and 1,046 miRNA biomarkers, taken from The Cancer Genome Atlas (TCGA) database.
E-learning platforms, such as Blackboard have two major shortcomings: limited data capture as a result of the limitations of SCORM (Shareable Content Object Reference Model), and lack of incorporation of Artificial Intelligence (AI) and machine learning algorithms which could lead to better course adaptations. With the recent development of Experience Application Programming Interface (xAPI), a large amount of additional types of data can be captured and that opens a window of possibilities from which online education can benefit. In a corporate setting, where companies invest billions on the learning and development of their employees, some learner behaviours can be troublesome for they can hinder the knowledge development of a learner. Behaviours that hinder the knowledge development also raise ambiguity about learner’s knowledge mastery, specifically those related to gaming the system. Furthermore, a company receives little benefit from their investment if employees are passing courses without possessing the required knowledge and potential compliance risks may arise. Using xAPI and rules derived from a state-of-the-art review, we identified three learner behaviours, primarily related to guessing, in a corporate compliance course. The identified behaviours are: trying each option for a question, specifically for multiple-choice questions; selecting a single option for all the questions on the test; and continuously repeating tests upon failing as opposed to going over the learning material. These behaviours were detected on learners who repeated the test at least 4 times before passing the course. These findings suggest that gauging the mastery of a learner from multiple-choice questions test scores alone is a naive approach. Thus, next steps will consider the incorporation of additional data points, knowledge estimation models to model knowledge mastery of a learner more accurately, and analysis of the data for correlations between knowledge development and identified learner behaviours. Additional work could explore how learner behaviours could be utilised to make changes to a course. For example, course content may require modifications (certain sections of learning material may be shown to not be helpful to many learners to master the learning outcomes aimed at) or course design (such as the type and duration of feedback).
Surface Mount Technology (SMT) is widely used in the area of the electronic assembly in which the electronic components are mounted to the surface of the printed circuit board (PCB). Most of the defects in the SMT process are mainly related to the quality of solder paste printing. These defects lead to considerable manufacturing costs in the electronics assembly industry. Therefore, the solder paste inspection (SPI) machine for controlling and monitoring the amount of solder paste printing has become an important part of the production process. So far, the setting of the SPI threshold is based on statistical analysis and experts’ experiences to determine the appropriate threshold settings. Because the production data are not normal distribution and there are various variations in the production processes, defects related to solder paste printing still occur. In order to solve this problem, this paper proposes an online machine learning algorithm, called the automatic threshold adjustment (ATA) algorithm, and closed-loop architecture in the SMT process to determine the best threshold settings. Simulation experiments prove that our proposed threshold settings improve the accuracy from 99.85% to 100%.
A key issue in stock investment is how to select representative features for stock selection. The objective of this paper is to firstly determine whether an automated stock investment system, using machine learning techniques, may be used to identify a portfolio of growth stocks that are highly likely to provide returns better than the stock market index. The second objective is to identify the technical features that best characterize whether a stock’s price is likely to go up and to identify the most important factors and their contribution to predicting the likelihood of the stock price going up. Unsupervised machine learning techniques, such as cluster analysis, were applied to the stock data to identify a cluster of stocks that was likely to go up in price – portfolio 1. Next, the principal component analysis technique was used to select stocks that were rated high on component one and component two – portfolio 2. Thirdly, a supervised machine learning technique, the logistic regression method, was used to select stocks with a high probability of their price going up – portfolio 3. The predictive models were validated with metrics such as, sensitivity (recall), specificity and overall accuracy for all models. All accuracy measures were above 70%. All portfolios outperformed the market by more than eight times. The top three stocks were selected for each of the three stock portfolios and traded in the market for one month. After one month the return for each stock portfolio was computed and compared with the stock market index returns. The returns for all three stock portfolios was 23.87% for the principal component analysis stock portfolio, 11.65% for the logistic regression portfolio and 8.88% for the K-means cluster portfolio while the stock market performance was 0.38%. This study confirms that an automated stock investment system using machine learning techniques can identify top performing stock portfolios that outperform the stock market.
This paper presents a Machine Learning (ML) approach to support Meningitis diagnosis in patients at a children’s hospital in Sao Paulo, Brazil. The aim is to use ML techniques to reduce the use of invasive procedures, such as cerebrospinal fluid (CSF) collection, as much as possible. In this study, we focus on predicting the probability of Meningitis given the results of a blood and urine laboratory tests, together with the analysis of pain or other complaints from the patient. We tested a number of different ML algorithms, including: Adaptative Boosting (AdaBoost), Decision Tree, Gradient Boosting, K-Nearest Neighbors (KNN), Logistic Regression, Random Forest and Support Vector Machines (SVM). Decision Tree algorithm performed best, with 94.56% and 96.18% accuracy for training and testing data, respectively. These results represent a significant aid to doctors in diagnosing Meningitis as early as possible and in preventing expensive and painful procedures on some children.
2016 has become the year of the Artificial Intelligence explosion. AI technologies are getting more and more matured that most world well-known tech giants are making large investment to increase the capabilities in AI. Machine learning is the science of getting computers to act without being explicitly programmed, and deep learning is a subset of machine learning that uses deep neural network to train a machine to learn features directly from data. Deep learning realizes many machine learning applications which expand the field of AI. At the present time, deep learning frameworks have been widely deployed on servers for deep learning applications in both academia and industry. In training deep neural networks, there are many standard processes or algorithms, but the performance of different frameworks might be different. In this paper we evaluate the running performance of two state-of-the-art distributed deep learning frameworks that are running training calculation in parallel over multi GPU and multi nodes in our cloud environment. We evaluate the training performance of the frameworks with ResNet-50 convolutional neural network, and we analyze what factors that result in the performance among both distributed frameworks as well. Through the experimental analysis, we identify the overheads which could be further optimized. The main contribution is that the evaluation results provide further optimization directions in both performance tuning and algorithmic design.
Attention-Deficit/Hyperactivity Disorder (ADHD), epilepsy, and autism affect millions of children worldwide, many of which are undiagnosed despite the fact that all of these disorders are detectable in early childhood. Late diagnosis can cause severe problems due to the late treatment and to the misconceptions and lack of awareness as a whole towards these disorders. Moreover, electroencephalography (EEG) has played a vital role in the assessment of neural function in children. Therefore, quantitative EEG measurement will be utilized as a tool for use in the evaluation of patients who may have ADHD, epilepsy, and autism. We propose a screening tool that uses EEG signals and machine learning algorithms to detect these disorders at an early age in an automated manner. The proposed classifiers used with epilepsy as a step taken for the work done so far, provided an accuracy of approximately 97% using SVM, Naïve Bayes and Decision tree, while 98% using KNN, which gives hope for the work yet to be conducted.
Cognitive decline and frailty is apparent in older adults leading to an increased likelihood of the risk of falling. Currently health care professionals have to make professional decisions regarding such risks, and hence make difficult decisions regarding the future welfare of the ageing population. This study uses health data from The Irish Longitudinal Study on Ageing (TILDA), focusing on adults over the age of 50 years, in order to analyse health risk factors and predict the likelihood of falls. This prediction is based on the use of machine learning algorithms whereby health risk factors are used as inputs to predict the likelihood of falling. Initial results show that health risk factors such as long-term health issues contribute to the number of falls. The identification of such health risk factors has the potential to inform health and social care professionals, older people and their family members in order to mitigate daily living risks.
Accurate prediction of NOx emission is a continuous challenge in the field of diesel engine-out emission modeling. Performing experiments for each conditions and scenario cost significant amount of money and man hours, therefore model-based development strategy has been implemented in order to solve that issue. NOx formation is highly dependent on the burn gas temperature and the O2 concentration inside the cylinder. The current empirical models are developed by calibrating the parameters representing the engine operating conditions with respect to the measured NOx. This makes the prediction of purely empirical models limited to the region where it has been calibrated. An alternative solution to that is presented in this paper, which focus on the utilization of in-cylinder combustion parameters to form a predictive semi-empirical NOx model. The result of this work is shown by developing a fast and predictive NOx model by using the physical parameters and empirical correlation. The model is developed based on the steady state data collected at entire operating region of the engine and the predictive combustion model, which is developed in Gamma Technology (GT)-Power by using Direct Injected (DI)-Pulse combustion object. In this approach, temperature in both burned and unburnt zone is considered during the combustion period i.e. from Intake Valve Closing (IVC) to Exhaust Valve Opening (EVO). Also, the oxygen concentration consumed in burnt zone and trapped fuel mass is also considered while developing the reported model. Several statistical methods are used to construct the model, including individual machine learning methods and ensemble machine learning methods. A detailed validation of the model on multiple diesel engines is reported in this work. Substantial numbers of cases are tested for different engine configurations over a large span of speed and load points. Different sweeps of operating conditions such as Exhaust Gas Recirculation (EGR), injection timing and Variable Valve Timing (VVT) are also considered for the validation. Model shows a very good predictability and robustness at both sea level and altitude condition with different ambient conditions. The various advantages such as high accuracy and robustness at different operating conditions, low computational time and lower number of data points requires for the calibration establishes the platform where the model-based approach can be used for the engine calibration and development process. Moreover, the focus of this work is towards establishing a framework for the future model development for other various targets such as soot, Combustion Noise Level (CNL), NO2/NOx ratio etc.
With the continuous increment of smart meter installations across the globe, the need for processing of the load data is evident. Clustering-based load profiling is built upon the utilization of unsupervised machine learning tools for the purpose of formulating the typical load curves or load profiles. The most commonly used algorithm in the load profiling literature is the K-means. While the algorithm has been successfully tested in a variety of applications, its drawback is the strong dependence in the initialization phase. This paper proposes a novel modified form of the K-means that addresses the aforementioned problem. Simulation results indicate the superiority of the proposed algorithm compared to the K-means.
The Greek Energy Market is structured as a mandatory pool where the producers make their bid offers in day-ahead basis. The System Operator solves an optimization routine aiming at the minimization of the cost of produced electricity. The solution of the optimization problem leads to the calculation of the System Marginal Price (SMP). Accurate forecasts of the SMP can lead to increased profits and more efficient portfolio management from the producer`s perspective. Aim of this study is to provide a comparative analysis of various machine learning models such as artificial neural networks and neuro-fuzzy models for the prediction of the SMP of the Greek market. Machine learning algorithms are favored in predictions problems since they can capture and simulate the volatilities of complex time series.
A major challenge in medical studies, especially those that are longitudinal, is the problem of missing measurements which hinders the effective application of many machine learning algorithms. Furthermore, recent Alzheimer's Disease studies have focused on the delineation of Early Mild Cognitive Impairment (EMCI) and Late Mild Cognitive Impairment (LMCI) from cognitively normal controls (CN) which is essential for developing effective and early treatment methods. To address the aforementioned challenges, this paper explores the potential of using the eXtreme Gradient Boosting (XGBoost) algorithm in handling missing values in multiclass classification. We seek a generalized classification scheme where all prodromal stages of the disease are considered simultaneously in the classification and decision-making processes. Given the large number of subjects (1631) included in this study and in the presence of almost 28% missing values, we investigated the performance of XGBoost on the classification of the four classes of AD, NC, EMCI, and LMCI. Using 10-fold cross validation technique, XGBoost is shown to outperform other state-of-the-art classification algorithms by 3% in terms of accuracy and F-score. Our model achieved an accuracy of 80.52%, a precision of 80.62% and recall of 80.51%, supporting the more natural and promising multiclass classification.
Gradient boosting methods have been proven to be a very important strategy. Many successful machine learning solutions were developed using the XGBoost and its derivatives. The aim of this study is to investigate and compare the efficiency of three gradient methods. Home credit dataset is used in this work which contains 219 features and 356251 records. However, new features are generated and several techniques are used to rank and select the best features. The implementation indicates that the LightGBM is faster and more accurate than CatBoost and XGBoost using variant number of features and records.
Machine Learning and Data Mining are the two important tools for extracting useful information and knowledge from large datasets. In machine learning, classification is a wildly used technique to predict qualitative variables and is generally preferred over regression from an operational point of view. Due to the enormous increase in air pollution in various countries especially China, Air Quality Classification has become one of the most important topics in air quality research and modelling. This study aims at introducing a hybrid classification model based on information theory and Support Vector Machine (SVM) using the air quality data of four cities in China namely Beijing, Guangzhou, Shanghai and Tianjin from Jan 1, 2014 to April 30, 2016. China's Ministry of Environmental Protection has classified the daily air quality into 6 levels namely Serious Pollution, Severe Pollution, Moderate Pollution, Light Pollution, Good and Excellent based on their respective Air Quality Index (AQI) values. Using the information theory, information gain (IG) is calculated and feature selection is done for both categorical features and continuous numeric features. Then SVM Machine Learning algorithm is implemented on the selected features with cross-validation. The final evaluation reveals that the IG and SVM hybrid model performs better than SVM (alone), Artificial Neural Network (ANN) and K-Nearest Neighbours (KNN) models in terms of accuracy as well as complexity.
Online marketplaces are not only digital places where consumers buy and sell merchandise, and they are also destinations for brands to connect with real consumers at the moment when customers are in the shopping mindset. For many marketplaces, brands have been important partners through advertising. There can be, however, a risk of advertising impacting a consumer’s shopping journey if it hurts the use experience or takes the user away from the site. Both could lead to the loss of transaction revenue for the marketplace. In this paper, we present user-based methods for cannibalization control by selectively turning off ads to users who are likely to be cannibalized by ads subject to business objectives. We present ways of measuring cannibalization of advertising in the context of an online marketplace and propose novel ways of measuring cannibalization through purchase propensity and uplift modeling. A/B testing has shown that our methods can significantly improve user purchase and engagement metrics while operating within business objectives. To our knowledge, this is the first paper that addresses cannibalization mitigation at the user-level in the context of advertising.