Predicting part quality early during an injection molding cycle

P125. Development of a novel ensemble machine learning algorithm for prediction of complications and readmission after anterior cervical spinal fusion

The Autoregressive Integrated Moving Average model (ARIMA) is a popular time-series model used to predict future trends in economics, energy markets, and stock markets. It has not been widely applied to enrollment forecasting in higher education. The accuracy of the ARIMA model heavily relies on the length of time series. Researchers and practitioners often utilize the most recent - to -years of historical data to predict future enrollment; however, the accuracy of enrollment projection under different lengths of time series has never been investigated and compared. A simulation and an empirical study were conducted to thoroughly investigate the accuracy of ARIMA forecasting under four different lengths of time series. When the ARIMA model completely captured the historical changing trajectories, it provided the most accurate predictions of student enrollment with 20-years of historical data and had the lowest forecasting accuracy with the shortest time series. The results of this paper contribute as a reference to studies in the enrollment projection and time-series forecasting. It provides a practical impact on enrollment strategies, budges plans, and financial aid policies at colleges and institutions across countries.

Assessment of the length of rainfall time series for rainwater harvesting in buildings

Industry 4.0 is generally defined as a development system that compels the digitalization of processes to create integrated and autonomous systems. The process tracking of parts is very important in terms of detecting missed faulty products. Some defects that escape from quality control directly affect the end-user. Machine learning algorithms have been used to predict changes in the quality control processes and defective products, toward real-time and effective data processing. Thus, the highest quality of the final product will be delivered to the customer and to reduce the defective production coming out of the manufacturing chain. In this article, the study aims to establish a predictive quality model that can detect defect-free approved but faulty products overlooked during the quality inspection operations. Machine learning methods are used to analyze the relationship between quality control data and customer complaints. For this purpose, we use the last quality stage data of an automobile manufacturer’s brake system from 2018 to 2020. Machine learning models are constructed using logistic regression, ridge regression, support vector machine, random forest classification tree, gradient boost, XGBoost, LightGBM, and CatBoost algorithms. The results of specificity and negative prediction value show that the Gradient Boost and CatBoost algorithms have the best classification benefit for detecting the rare events.KeywordsQuality 4.0Industry 4.0Predictive qualityAutomobile industryMachine learningFault detectionRare event detection

In analyzing the Human immunodeficiency virus (HIV) epidemic dynamics, the biggest problem is uncertainty when planning for the future. In future evaluations, predicting what might happen will make the decisions’ results more realistic. Policymakers will have the opportunity to take precautions against any negative changes that may occur. Machine learning methods that produce good and effective predictive results are needed to plan future policies, eliminate the negativities and overcome deciding in an uncertain environment. In this study, seven machine learning models used to make time-series analysis for medical purposes are theoretically explained. Machine learning methods such as Linear Regression, RepTree, Alternating Model Trees, M5, k Nearest Neighbor (kNN), Autoregressive Integrated Moving Average (ARIMA), and Random Forest were used. The dynamics of the HIV epidemic in Turkey have been made stationary time series, considering compliance of the correlation. Then, the time series were preprocessed using the Moving Average technique, and the time series was softened. The time series is divided into 2/3 training and 1/3 test sets. Machine learning methods were trained using these sets, parameter optimization of models was made and tested. Then these models were used to forecast the HIV epidemic Dynamics in Turkey in 3 years between 2019-Q4 and 2022-Q3. The Random Forest method has been successful as the model that produces the least error rate (Mean Absolute Percentage Error, MAPE) among these seven models. According to the estimation results of the Random Forest model, R2 (the coefficient of determination) value was 82.16%, E (efficiency) value was 0.6268, Slope value was 2.3362, and MAPE value was 5.4132%. The Random Forest model has been observed to give excellent results for the three-year forecast of dynamics of the HIV epidemic in Turkey.

Measurement of time series complexity and predictability is sometimes the cornerstone for proposing solutions to topology and congestion control problems in sensor networks. As a method of measuring time series complexity and predictability, multiscale entropy (MSE) has been widely applied in many fields. However, sample entropy, which is the fundamental component of MSE, measures the similarity of two subsequences of a time series with either zero or one, but without in-between values, which causes sudden changes of entropy values even if the time series embraces small changes. This problem becomes especially severe when the length of time series is getting short. For solving such the problem, we propose flexible multiscale entropy (FMSE), which introduces a novel similarity function measuring the similarity of two subsequences with full-range values from zero to one, and thus increases the reliability and stability of measuring time series complexity. The proposed method is evaluated on both synthetic and real time series, including white noise, 1/f noise and real vibration signals. The evaluation results demonstrate that FMSE has a significant improvement in reliability and stability of measuring complexity of time series, especially when the length of time series is short, compared to MSE and composite multiscale entropy (CMSE). The proposed method FMSE is capable of improving the performance of time series analysis based topology and traffic congestion control techniques.

Performance evaluation of machine learning techniques in predicting cumulative absolute velocity

Throughout recent years, dynamic time warping (DTW) has remained as a robust similarity measure in time series classification (TSC). 1-nearest neighbor (1-NN) algorithm with DTW is the most widely used classification method on time series serving as a benchmark. With the increasing demand for TSC on low-resource devices and the widespread of wearable devices, the need for a efficient and accurate time series classifier has never been higher. Although 1-NN DTW attains accurate results, it highly falls back on efficiency due to its quadratic complexity in the length of time series. In this paper, we propose a new approximation method for reducing the length of the time series as the input of DTW. We call it control chart approximation (CCA), after a similar concept used in statistical quality control processing. CCA representation approximates raw time series by transforming them into a set of segments with aggregated values and durations forming a reduced 3-D vector. We also propose an adaptation of DTW in 3-D space as a distance measure for 1-NN classifier, and denote the method as 1-NN 3-D DTW. Our experiments on 85 datasets from UCR archive—including 28 long-length (>500 points) time series datasets—show up to two orders of magnitude performance gain in running time compared to the state-of-the-art 1-NN DTW implementation. Moreover, it shows similar or better accuracy on the long time series in the experiment.

To develop a model for predicting biochemical recurrence (BCR) in patients with long follow-up periods using clinical parameters and the machine learning (ML) methods. Patients who underwent robot-assisted radical prostatectomy between January 2014 and December 2019 were retrospectively reviewed. Patients who did not have BCR were assigned to Group 1, while those diagnosed with BCR were assigned to Group 2. The patient's demographic data, preoperative and postoperative parameters were all recorded in the database. Three different ML algorithms were employed: random forest, K-nearest neighbour, and logistic regression. Three hundred and sixty-eight patients were included in this study. Among these patients, 295 (80.1%) did not have BCR (Group 1), while 73 (19.8%) had BCR (Group 2). The mean duration of follow-up and duration until the diagnosis of BCR was calculated as 35.2 ± 16.7 and 11.5 ± 11.3 months, respectively. The multivariate analysis revealed that NLR, PSAd, risk classification, PIRADS score, T stage, presence or absence of positive surgical margin, and seminal vesicle invasion were predictive for BCR. Classic Cox regression analysis had an area under the curve (AUC) of 0.915 with a sensitivity and specificity of 90.6% and 79.8%. The AUCs for receiver-operating characteristic curves for random forest, K nearest neighbour, and logistic regression were 0.95, 0.93, and 0.93, respectively. All ML models outperformed the conventional statistical regression model in the prediction of BCR after prostatectomy. The construction of more reliable and potent models will provide the clinicians and patients with advantages such as more accurate risk classification, prognosis estimation, early intervention, avoidance of unnecessary treatments, relatively lower morbidity and mortality. The ML methods are cheap, and their powers increase with increasing data input; we believe that their clinical use will increase over time.

Missing data are critical deficiency in the investigation of displacement measurement in urban Internet of Things system. In the insight of recovering missing displacement data, this article presents a data-driven and high-dimensional gap-imputation method, Tucker decomposition with L2 regularization. Results on the global navigation satellite system (GNSS) time series collected from an intelligent structural health monitoring system show that the recovery accuracy is improved compared with some popular benchmark methods. When the missing rate is 50%, compared with singular spectrum analysis, singular value decomposition, CP optimization algorithm, k -nearest neighbors, and Tucker decomposition via alternating least squares, Tucker decomposition with L2 regularization can improve the average mean absolute error by about 4.74, 4.95, 5.82, 2.29, and 5.67 mm for all locations. It can be concluded that the consideration of multiple temporal correlations is necessary for missing data imputation. Compared with matrix decomposition, tensor decomposition can improve the ability for high-dimensional correlations in the GNSS time series.

In the era of artificial intelligence, machine learning methods are successfully used in various fields. Machine learning has attracted extensive attention from investors in the financial market, especially in stock price prediction. However, one argument for the machine learning methods used in stock price prediction is that they are black-box models which are difficult to interpret. In this paper, we focus on the future stock price prediction with the historical stock price by machine learning and deep learning methods, such as support vector machine (SVM), random forest (RF), Bayesian classifier (BC), decision tree (DT), multilayer perceptron (MLP), convolutional neural network (CNN), bi-directional long-short term memory (BiLSTM), the embedded CNN, and the embedded BiLSTM. Firstly, we manually design several financial time series where the future price correlates with the historical stock prices in pre-designed modes, namely the curve-shape-feature (CSF) and the non-curve-shape-feature (NCSF) modes. In the CSF mode, the future prices can be extracted from the curve shapes of the historical stock prices. Conversely, in the NCSF mode, they can’t. Secondly, we apply various algorithms to those pre-designed and real financial time series. We find that the existing machine learning and deep learning algorithms fail in stock price prediction because in the real financial time series, less information of future prices is contained in the CSF mode, and perhaps more information is contained in the NCSF. Various machine learning and deep learning algorithms are good at handling the CSF in historical data, which are successfully applied in image recognition and natural language processing. However, they are inappropriate for stock price prediction on account of the NCSF. Therefore, accurate stock price prediction is the key to successful investment, and new machine learning algorithms handling the NCSF series are needed.

In view of the current status of the university funding system, combined with the research results of precise funding of universities, and using machine learning methods as technical support, construct a model for accurately identifying poor families. Based on the poverty dataset of Costa Rica, we analyzed and processed it, then constructed a sample database, and used machine learning methods to design and implement a model for accurately identifying poor families. In the process of model verification, this paper uses machine learning algorithms such as logistic regression, support vector machines, K nearest neighbors, decision trees, and random forests to identify poor households. The experimental results show that the performance of the integrated machine learning algorithm is better than that of the traditional machine learning algorithm, and the prediction performance of the decision tree algorithm is the best, with an average accuracy rate of 89%. The research results of this paper can realize the analysis and prediction of poor family data, and then accurately identify poor students, and realize the new model of differentiated and precise funding based on people.

<p>Merging data from different instruments is required to construct long time data records of soil moisture (SM). This is the goal of projects such as the ESA Climate Change Initiative (CCI) for SM (Gruber et al., 2019), which uses both active and passive microwave sensors. Currently, the GLDAS v2.1 model is used as reference to re-scale active and passive time series by matching their Cumulative Density Function (CDF) to that of the model. Removing the dependency on models is important, in particular for data assimilation applications into hydrological or climate models, and it has been proposed (Van der Schalie et al., 2018) to use L-band data from one of the two instruments specifically designed to measure SM, ESA Soil Moisture and Ocean Salinity (SMOS) and NASA Soil Moisture Active Passive (SMAP) satellites, as reference to re-scale other time series.<br>To investigate this approach, AMSR-2 SM time series obtained from C1-, C2- and X-band observations using LPRM (Land Parameter Retrieval Model) were re-scaled by CDF-matching (Brocca et al., 2011) using different SMAP and SMOS official (SMAP L2 V005, SMOS L3 V300, SMOS NRT V100&V200) and research (SMOS IC V103) SM products as well as the SMAP and SMOS LPRM v6 SM data used by the ESA CCI. The time series re-scaled using L-band remote sensing data were compared to those re-scaled using GLDAS and were evaluated against in situ measurements at several hundred sites retrieved from the International Soil Moisture Network (Dorigo et al., 2011). The results were analyzed as a function of the land cover class and the Koppen-Geiger climate classification.<br>Overall, AMSR-2 time series re-scaled using SMAP L2, SMAP LPRM and SMOS IC data sets as reference gave the best correlations with respect to in situ measurements, similar to those obtained by the time series re-scaled using GLDAS and slightly better than those of the original AMSR-2 time series. These results imply that different SMAP and SMOS products could actually be used to replace GLDAS as reference for the re-scaling of other sensors time series within the ESA CCI. However, one must bear in mind that this study is limited to the re-scaling of AMSR-2 data at a few hundred sites.<br>For a more detailed assessment of the L-band data set to be used for a global re-scaling, it is necessary to investigate other effects such as the spatial coverage or the time series length. SMAP spatial coverage is better than that of SMOS in regions affected by radio frequency interference. In contrast, the length of SMAP time series can be too short to capture the long term SM variability for climate applications in some regions. The CDF of SMOS time series computed from the date of SMAP launch is significantly different to those of the full length SMOS time series in some regions of the Globe. Possible ways of using a coherent SMAP/SMOS L-band data set will be discussed.</p>

Coevolving multiple time series are ubiquitous and naturally appear in a variety of high-impact applications, ranging from environmental monitoring, computer network traffic monitoring, motion capture, to physiological signal in health care and many more. In many scenarios, the multiple time series data is often accompanied by some contextual information in the form of networks. In this paper, we refer to such multiple time series, together with its embedded network as a network of coevolving time series. In order to unveil the underlying patterns of a network of coevolving time series, we propose DCMF, a dynamic contextual matrix factorization algorithm. The key idea is to find the latent factor representation of the input time series and that of its embedded network simultaneously. Our experimental results on several real datasets demonstrate that our method (1) outperforms its competitors, especially when there are lots of missing values; and (2) enjoys a linear scalability w.r.t. the length of time series.

Diabetic Retinopathy (DR) is a highly threatening microvascular complication of diabetes mellitus. Diabetic patients must be screened annually for DR; however, it is practically not viable due to the high volume of patients, lack of resources, economic burden, and cost of the screening procedure. The use of machine learning (ML) classifiers in medical science is an emerging frontier and can help in assisted diagnosis. The few available proposed models perform best when used in similar population cohorts and their external validation has been questioned. Therefore, the purpose of our research is to classify the DR using different ML methods on Saudi diabetic data, propose the best method based on accuracy and identify the most discriminative interpretable features using the socio-demographic and clinical information. This cross-sectional study was conducted among 327 diabetic patients in Almajmaah, Saudi Arabia. Socio-demographic and clinical data were collected using a systematic random sampling technique. For DR classification, ML algorithm including, linear discriminant analysis, support vector machine, K nearest neighbor, random forest and its variate ranger random forest classifiers were used through cross-validation resampling procedure. In classifying DR, ranger random forest outperforms the other methods by accurately classifying 86% of the DR patients on the test data. HbA1c (p<0.001) and duration of diabetes (p<0.001) were the most influential risk factor that best discriminated the DR patients. Other influential risk factors were the body mass index (p<0.001), age-onset (p<0.001), age (p<0.001), systolic blood pressure (p<0.05), and the use of medication (p<0.05) that significantly discriminated the DR patients. Based on the present study findings, integrating ophthalmology and ML can transform diagnosing the disease pattern that can help generate a compelling clinical effect. ML can be used as an added tool for clinical decision-making and must not be the sole substitute for a clinician. We will work to examine the classification performance of multi-class data using more sophisticated ML methods.