Multilayer Artificial Neural Network Research Articles

Machine learning as a tool for diagnostic and prognostic research in coronary artery disease

Machine learning (ML) are the central tool of artificial intelligence, the use of which makes it possible to automate the processing and analysis of large data, reveal hidden or non-obvious patterns and learn a new knowledge. The review presents an analysis of literature on the use of ML for diagnosing and predicting the clinical course of coronary artery disease. We provided information on reference databases, the use of which allows to develop models and validate them (European ST-T Database, Cleveland Heart Disease database, Multi-Ethnic Study of Atherosclerosis, etc.). The advantages and disadvantages of individual ML methods (logistic regression, support vector machines, decision trees, naive Bayesian classifier, k-nearest neighbors) for the development of diagnostic and predictive algorithms are shown. The most promising ML methods include deep learning, which is implemented using multilayer artificial neural networks. It is assumed that the improvement of ML-based models and their introduction into clinical practice will help support medical decision-making, increase the effectiveness of treatment and optimize health care costs.

Mutual Information and Meta-Heuristic Classifiers Applied to Bearing Fault Diagnosis in Three-Phase Induction Motors

Three-phase induction motors are extensively used in industrial processes due to their robustness, adaptability to different operating conditions, and low operation and maintenance costs. Induction motor fault diagnosis has received special attention from industry since it can reduce process losses and ensure the reliable operation of industrial systems. Therefore, this paper presents a study on the use of meta-heuristic tools in the diagnosis of bearing failures in induction motors. The extraction of the fault characteristics is performed based on mutual information measurements between the stator current signals in the time domain. Then, the Artificial Bee Colony algorithm is used to select the relevant mutual information values and optimize the pattern classifier input data. To evaluate the classification accuracy under various levels of failure severity, the performance of two different pattern classifiers was compared: The C4.5 decision tree and the multi-layer artificial perceptron neural networks. The experimental results confirm the effectiveness of the proposed approach.

Impact of COVID-19 on GDP of major economies: Application of the artificial neural network forecaster

The ongoing COVID-19 pandemic has caused global health impacts, and governments have restricted movements to a certain extent. Such restrictions have led to disruptions in economic activities. In this paper, the GDP figures for the April–June quarter of 2020 for eight countries, namely, the United States, Mexico, Germany, Italy, Spain, France, India, and Japan, are forecasted. Considering that artificial neural network models have higher forecasting accuracy than statistical methods, a multilayer artificial neural network model is developed in this paper. This model splits the dataset into two parts: the first with 80% of the observations and the second with 20%. The model then uses the first part to optimize the forecasting accuracy and then applies the optimized parameters to the second part of the dataset to assess the model performance. A forecasting error of less than 2% is achieved by the model during the testing procedure. The forecasted GDP figures show that the April–June quarter of the current year experienced sharp declines in GDP for all countries. Moreover, the annualized GDP growth is expected to reach double-digit negative growth rates. Such alarming prospects require urgent rescue actions by governments.

Artificial intelligence prediction of the effect of rehabilitation in whiplash associated disorder

The active cervical range of motion (aROM) is assessed by clinicians to inform their decision-making. Even with the ability of neck motion to discriminate injured from non-injured subjects, the mechanisms to explain recovery or persistence of WAD remain unclear. There are few studies of ROM examinations with precision tools using kinematics as predictive factors of recovery rate. The present paper will evaluate the performance of an artificial neural network (ANN) using kinematic variables to predict the overall change of aROM after a period of rehabilitation in WAD patients. To achieve this goal the neck kinematics of a cohort of 1082 WAD patients (55.1% females), with mean age 37.68 (SD 12.88) years old, from across Spain were used. Prediction variables were the kinematics recorded by the EBI® 5 in routine biomechanical assessments of these patients. These include normalized ROM, speed to peak and ROM coefficient of variation. The improvement of aROM was represented by the Neck Functional Holistic Analysis Score (NFHAS). A supervised multi-layer feed-forward ANN was created to predict the change in NFHAS. The selected architecture of the ANN showed a mean squared error of 308.07–272.75 confidence interval for a 95% in the Monte Carlo cross validation. The performance of the ANN was tested with a subsample of patients not used in the training. This comparison resulted in a medium correlation with R = 0.5. The trained neural network to predict the expected difference in NFHAS between baseline and follow up showed modest results. While the overall performance is moderately correlated, the error of this prediction is still too large to use the method in clinical practice. The addition of other clinically relevant factors could further improve prediction performance.

Crack classification in rotor-bearing system by means of wavelet transform and deep learning methods: an experimental investigation

Parallel with significant growth in industry, especially mysteries related to energy engineering, condition monitoring of rotating systems have been experiencing a noticeable increase. One of the prevalent faults in these systems is fatigue crack, so finding reliable procedures in identification of cracks in rotating shafts has become a pressing problem among engineers during recent decades. While a vast majority of cracked rotors can operate for a specific period of time, to prevent catastrophic failures, crack detection and measuring its characteristics (i.e. size and its location) seem to be essential. In the present essay, a hybrid procedure, consisting of Deep Learning and Discrete Wavelet transform (DWT), is applied in detection of a breathing transverse crack and its depth in a rotor-bearing-disk system. DWT with Daubechies 32(db32) as wavelet mother function is applied in signal noise reduction until level 6, also its Relative Wavelet Energy (RWE) and Wavelet entropy (WE) are extracted. A characteristic vector that is a combination of RWE and WE is considered as input to a multi-layer Artificial Neural Network (ANN). In this supervised learning classifier, a multi-layer Perceptron neural network is used; in addition, Rectified Linear Unit (ReLU) function is exerted as activation function in both hidden and output layers. By comparing the results, it can be seen that the applied procedure has strong capacity in identification of crack and its size in the rotor system.

Radiomics of high-resolution computed tomography for the differentiation between cholesteatoma and middle ear inflammation: effects of post-reconstruction methods in a dual-center study

ObjectivesTo evaluate the performance of radiomic features extracted from high-resolution computed tomography (HRCT) for the differentiation between cholesteatoma and middle ear inflammation (MEI), and to investigate the impact of post-reconstruction harmonization and data resampling.MethodsOne hundred patients were included in this retrospective dual-center study: 48 with histology-proven cholesteatoma (center A: 23; center B: 25) and 52 with MEI (A: 27; B: 25). Radiomic features (co-occurrence and run-length matrix, absolute gradient, autoregressive model, Haar wavelet transform) were extracted from manually defined 2D-ROIs. The ten best features for lesion differentiation were selected using probability of error and average correlation coefficients. A multi-layer perceptron feed-forward artificial neural network (MLP-ANN) was used for radiomics-based classification, with histopathology serving as the reference standard (70% of cases for training, 30% for validation). The analysis was performed five times each on (a) unmodified data and on data that were (b) resampled to the same matrix size, and (c) corrected for acquisition protocol differences using ComBat harmonization.ResultsUsing unmodified data, the MLP-ANN classification yielded an overall median area under the receiver operating characteristic curve (AUC) of 0.78 (0.72–0.84). Using original data from center A and resampled data from center B, an overall median AUC of 0.88 (0.82–0.99) was yielded, while using ComBat harmonized data, an overall median AUC of 0.89 (0.79–0.92) was revealed.ConclusionRadiomic features extracted from HRCT differentiate between cholesteatoma and MEI. When using multi-centric data obtained with differences in CT acquisition parameters, data resampling and ComBat post-reconstruction harmonization clearly improve radiomics-based lesion classification.Key Points• Unenhanced high-resolution CT coupled with radiomics analysis may be useful for the differentiation between cholesteatoma and middle ear inflammation.• Pooling of data extracted from inhomogeneous CT datasets does not appear meaningful without further post-processing.• When using multi-centric CT data obtained with differences in acquisition parameters, post-reconstruction harmonization and data resampling clearly improve radiomics-based soft-tissue differentiation.

Solvent Selection for Mitsunobu Reaction Driven by an Active Learning Surrogate Model

Optimization of a synthetic reaction with respect to solvent choice and operating conditions was implemented as a machine learning-based workflow. The approach is exemplified on the case study of selection of a promising solvent to maximize the yield of a Mitsunobu reaction producing isopropyl benzoate. A solvent was defined with 15 molecular descriptors, and a library of solvent descriptors was built. The descriptors were converted into a reduced dimensionality form using an Autoencoder. Experimental yields were used to train a multilayered artificial neural network (ANN) surrogate model, which was used for the optimization and design of experiments (DoE). DoE was performed in an active learning mode to reduce the number of experiments required for reaction optimization. The final surrogate model identified 1-chloropentane as a promising solvent, which resulted in an experimental yield of 93%.

NIR Spectroscopy for Discriminating and Predicting the Sensory Profile of Dry-Cured Beef "Cecina".

For Protected Geographical Indication (PGI)-labeled products, such as the dry-cured beef meat “cecina de León”, a sensory analysis is compulsory. However, this is a complex and time-consuming process. This study explores the viability of using near infrared spectroscopy (NIRS) together with artificial neural networks (ANN) for predicting sensory attributes. Spectra of 50 samples of cecina were recorded and 451 reflectance data were obtained. A feedforward multilayer perceptron ANN with 451 neurons in the input layer, a number of neurons varying between 1 and 30 in the hidden layer, and a single neuron in the output layer were optimized for each sensory parameter. The regression coefficient R squared (RSQ > 0.8 except for odor intensity) and mean squared error of prediction (MSEP) values obtained when comparing predicted and reference values showed that it is possible to predict accurately 23 out of 24 sensory parameters. Although only 3 sensory parameters showed significant differences between PGI and non-PGI samples, the optimized ANN architecture applied to NIR spectra achieved the correct classification of the 100% of the samples while the residual mean squares method (RMS-X) allowed 100% of non-PGI samples to be distinguished.

Assessment of a Machine-Learnt Adaptive Wall-Function in a Compressor Cascade With Sinusoidal Leading Edge

Abstract Near-wall modeling is one of the most challenging aspects of computational fluid dynamic computations. In fact, integration-to-the-wall with low-Reynolds approach strongly affects accuracy of results, but strongly increases the computational resources required by the simulation. A compromise between accuracy and speed to solution is usually obtained through the use of wall functions (WFs), especially in Reynolds averaged Navier–Stokes computations, which normally require that the first cell of the grid to fall inside the log-layer (50 < y+ < 200) (Wilcox, D. C., 1998, Turbulence Modeling for CFD, Vol. 2, DCW Industries, La Cañada, CA). This approach can be generally considered as robust, however the derivation of wall functions from attached flow boundary layers can mislead to nonphysical results in presence of specific flow topologies, e.g., recirculation, or whenever a detailed boundary layer representation is required (e.g., aeroacoustics studies) (Craft, T., Gant, S., Gerasimov, A., Lacovides, H., and Launder, B., 2002, “Wall – Function Strategies for Use in Turbulent Flow CFD,” Proceedings to 12th International Heat Transfer Conference, Grenoble, France, Aug. 18–23). In this work, a preliminary attempt to create an alternative data-driven wall function is performed, exploiting artificial neural networks (ANNs). Whenever enough training examples are provided, ANNs have proven to be extremely powerful in solving complex nonlinear problems (Goodfellow, I., Bengio, Y., Courville, A., and Bengio, Y., 2016, Deep Learning, Vol. 1, MIT Press, Cambridge, MA). The learner that is derived from the multilayer perceptron ANN, is here used to obtain two-dimensional, turbulent production and dissipation values near the walls. Training examples of the dataset have been initially collected either from large eddy simulation (LES) simulations of significant 2D test cases or have been found in open databases. Assessments on the morphology and the ANN training can be found in the paper. The ANN has been implemented in a Python environment, using scikit-learn and tensorflow libraries (Scikit-Learn Developers, 2019, “Scikit-learn v0.20.0 User Guide,” Software, Scikit-Learn Developers; Abadi, M., Barham, P., Chen, J., Chen, Z., Davis, A., Dean, J., Devin, M., Ghemawat, S., Irving, G., Isard, M., Kudlur, M., Levenberg, J., Monga, R., Moore, S., Murray, D. G., Steiner, B., Tucker, P., Vasudevan, V., Warden, P., Wicke, M., Yu, Y., and Zheng, X., 2016, “TensorFlow: A System for Large-Scale Machine Learning,” 12th Symposium on Operating Systems Design and Implementation, Savannah, GA, Nov. 2–4, pp. 265–283). The derived wall function is implemented in openfoam v-17.12 (CFD Direct, 2020, “OpenFoam User Guide v5,” CFD Direct, Caversham, UK), embedding the forwarding algorithm in run-time computations exploiting Python3.6m C_Api library. The data-driven wall function is here applied to k-epsilon simulations of a 2D periodic hill with different computational grids and to a modified compressor cascade NACA aerofoil with sinusoidal leading edge. A comparison between ANN enhanced simulations, available data and standard modelization is here performed and reported.

Prediction and control of surface roughness for the milling of Al/SiC metal matrix composites based on neural networks

In recent years, there has been a significant increase in the utilization of Al/SiC particulate composite materials in engineering fields, and the demand for accurate machining of such composite materials has grown accordingly. In this paper, a feed-forward multi-layered artificial neural network (ANN) roughness prediction model, using the Levenberg-Marquardt backpropagation training algorithm, is proposed to investigate the mathematical relationship between cutting parameters and average surface roughness during milling Al/SiC particulate composite materials. Milling experiments were conducted on a computer numerical control (CNC) milling machine with polycrystalline diamond (PCD) tools to acquire data for training the ANN roughness prediction model. Four cutting parameters were considered in these experiments: cutting speed, depth of cut, feed rate, and volume fraction of SiC. These parameters were also used as inputs for the ANN roughness prediction model. The output of the model was the average surface roughness of the machined workpiece. A successfully trained ANN roughness prediction model could predict the corresponding average surface roughness based on given cutting parameters, with a 2.08% mean relative error. Moreover, a roughness control model that could accurately determine the corresponding cutting parameters for a specific desired roughness with a 2.91% mean relative error was developed based on the ANN roughness prediction model. Finally, a more reliable and readable analysis of the influence of each parameter on roughness or the interaction between different parameters was conducted with the help of the ANN prediction model.

Application of Artificial Neural Networks to Assess the Mycological State of Bulk Stored Rapeseeds

Artificial neural networks (ANNs) constitute a promising modeling approach that may be used in control systems for postharvest preservation and storage processes. The study investigated the ability of multilayer perceptron and radial-basis function ANNs to predict fungal population levels in bulk stored rapeseeds with various temperatures (T = 12–30 °C) and water activity in seeds (aw = 0.75–0.90). The neural network model input included aw, temperature, and time, whilst the fungal population level was the model output. During the model construction, networks with a different number of hidden layer neurons and different configurations of activation functions in neurons of the hidden and output layers were examined. The best architecture was the multilayer perceptron ANN, in which the hyperbolic tangent function acted as an activation function in the hidden layer neurons, while the linear function was the activation function in the output layer neuron. The developed structure exhibits high prediction accuracy and high generalization capability. The model provided in the research may be readily incorporated into control systems for postharvest rapeseed preservation and storage as a support tool, which based on easily measurable on-line parameters can estimate the risk of fungal development and thus mycotoxin accumulation.

Artificial Intelligence, Accelerated in Parallel Computing and Applied to Nonintrusive Appliance Load Monitoring for Residential Demand-Side Management in a Smart Grid: A Comparative Study

A smart grid is a promising use-case of AIoT (AI (artificial intelligence) across IoT (internet of things)) that enables bidirectional communication among utilities that arises with demand response (DR) schemes for demand-side management (DSM) and consumers that manage their power demands according to received DR signals. Disaggregating composite electric energy consumption data from a single minimal set of plug-panel current and voltage sensors installed at the electric panel in a practical field of interest, nonintrusive appliance load monitoring (NIALM), a cost-effective load disaggregation approach for (residential) DSM, is able to discern individual electrical appliances concerned without accessing each of them by individual plug-load power meters (smart plugs) deployed intrusively. The most common load disaggregation approaches are based on machine learning algorithms such as artificial neural networks, while approaches based on evolutionary computing, metaheuristic algorithms considered as global optimization and search techniques, have recently caught the attention of researchers. This paper presents a genetic algorithm, developed in consideration of parallel evolutionary computing, and aims to address NIALM, whereby load disaggregation from composite electric energy consumption data is declared as a combinatorial optimization problem and is solved by the algorithm. The algorithm is accelerated in parallel, as it would involve large amounts of NIALM data disaggregated through evolutionary computing, chromosomes, and/or evolutionary cycles to dominate its performance in load disaggregation and excessively cost its execution time. Moreover, the evolutionary computing implementation based on parallel computing, a feed-forward, multilayer artificial neural network that can learn from training data across all available workers of a parallel pool on a machine (in parallel computing) addresses the same NIALM/load disaggregation. Where, a comparative study is made in this paper. The presented methodology is experimentally validated by and applied on a publicly available reference dataset.

Parameterized nonintrusive reduced‐order model for general unsteady flow problems using artificial neural networks

AbstractA nonintrusive reduced‐order model for nonlinear parametric flow problems is developed. It is based on extracting a reduced‐order basis from high‐order snapshots via proper orthogonal decomposition and using multilayered feedforward artificial neural networks to approximate the reduced‐order coefficients. The model is a generic and efficient approach for the reduction of time‐dependent parametric systems, including those described by partial differential equations. Since it is nonintrusive, it is independent of the high‐order computational method and can be used together with black‐box solvers. Numerical studies are presented for steady‐state isentropic nozzle flow with geometric parameterization and unsteady parameterized viscous Burgers equation. An adaptive sampling strategy is proposed to increase the quality of the neural network approximation while minimizing the required number of parameter samples and, as a direct consequence, the number of high‐order snapshots and the size of the network training set. Results confirm the accuracy of the nonintrusive approach as well as the speed‐up achieved compared with intrusive hyper‐reduced‐order approaches.

Smart Noise Reduction in SPR Sensors Response Using Multiple-ANN Design

Surface plasmon resonance (SPR) based sensors are commonly applied to detect molecular interactions from features extracted of the so-called SPR curve. The SPR curve represents the position of the resonance occurrence, and reflects the quality of the sensor measurements. The instrumentation used in the SPR sensors require very specific coupling conditions for optical-electronic, mechanical, and fluidic components to properly excitate the surface plasmons. Noise sources inherent to the sensor instrumentation are reflected in the SPR curve and affect the quality of its extracted features. A smart-filter based on multilayer artificial neural networks is designed to eliminate the main instrumental noise sources and noise due to sensing proceedings at the SPR sensor usage, which are perceived as distortions in the SPR curves. The filter, called NNSF, is embedded to SPR sensors operating in both spectral (WIM) and angular (AIM) interrogation modes, and achieved responses with a mean square error of 10−5, and tested model correlation indexes over 90%. The filter redressed noises, including artifacts such as metal-layer roughness, light source input angle/wavelength disagreements, prism birefringence and general oscillations in amplitude and phase of the curve due to no-rigorous control process in the environment and experimental routines. Thus, the NNSF delivers curves with enhanced features and consequently increases the sensor performances in the monitoring of the resonance features (Width, Energy, Phase, Asymmetry, and minimum Position). Furthermore, the proposed filter avoids replacement or incorporation of new components that cause considerable instrumental maintenance cost.

Water quality and macrophytes in the Danube River: Artificial neural network modelling

Ecological assessment of large rivers such as the Danube is a challenging task. Eutrophication was reported as one of the main drivers that structure aquatic communities in the Danube basin. Due to their sedentary nature, relatively slow growth/ long life spans, and engineering role in aquatic ecosystems, macrophytes are widely used in the detection of nutrient enrichment. In this study, macrophyte presence-absence data within the 3 km long reaches obtained from the Joint Danube Survey (JDS3) were used to predict the water quality of the Danube river and its main tributaries. For each water quality variable (dissolved oxygen, nitrate-nitrogen, and orthophosphates), a multi-layer feed-forward artificial neural network model (ANN) was constructed using the macrophytes as explanatory variables. Despite the limited number of samples (123) along the wide trophic gradient of the Danube, the model showed good predictive performances for the main river channel. The highest discrepancy between observed and predicted water quality was obtained for the samples collected in the tributaries or downstream from the tributaries' mouth, where the model predicted better trophic conditions compared to measured ones. From 64 analysed macrophyte species, 28 were selected by sensitivity analysis as key water quality indicators (KIS) for at least one environmental variable. KIS mainly belonged to the eutrophic tolerant submerged or emerged species with broad ecological amplitude, which reflects the significance of the developed model for use on rivers subjected to nutrient pollution. However, the use of the developed predictive model is restricted to the river sections having a water velocity suitable for macrophytes growth. The developed ANN architecture represents the modelling approach which could be applied to other lotic systems and biological quality elements.

Sensor Failure Management in Liquid Rocket Engine using Artificial Neural Network

This paper presents a novel Artificial Neural Network based Fault Detection, Isolation and Substitution (ANN-FDIS)algorithm for faulty sensor measurement in Liquid Rocket Engine (LRE). Fault detection and isolation are done by residual and fault flag logics and the trained multilayer perceptron model Artificial Neural Network (ANN) substitutes faulty sensor measurement. Data for ANN training, testing and validation are extracted from qualification and validation hot tests of LRE. Regression (R) and Mean Square Error (MSE) are considered for evaluating the ANN. During validation of this study, the faulty sensor is identified, isolated and data substituted from other input parameters with an error less than ±0.7%. This unique scheme does not require accurate modeling of the complicated LRE as well as sensor hardware redundancy which adds weight, space and power to rockets.

Simulating a combined lysis-cryptic and biological nitrogen removal system treating domestic wastewater at low C/N ratios using artificial neural network

In this study, a combined alkaline (ALK) and ultrasonication (ULS) sludge lysis-cryptic pretreatment and anoxic/oxic (AO) system (AO + ALK/ULS) was developed to enhance biological nitrogen removal (BNR) in domestic wastewater with a low carbon/nitrogen (C/N) ratio. A real-time control strategy for the AO + ALK/ULS system was designed to optimize the sludge lysate return ratio (RSLR) under variable sludge concentrations and variations in the influent C/N (⩽ 5). A multi-layered backpropagation artificial neural network (BPANN) model with network topology of 1 input layer, 3 hidden layers, and 1 output layer, using the Levenberg–Marquardt algorithm, was developed and validated. Experimental and predicted data showed significant concurrence, verified with a high regression coefficient (R2 = 0.9513) and accuracy of the BPANN. The BPANN model effectively captured the complex nonlinear relationships between the related input variables and effluent output in the combined lysis-cryptic + BNR system. The model could be used to support the real-time dynamic response and process optimization control to treat low C/N domestic wastewater.

Analysis of Forming Behavior in Cold Forging of AISI 1010 Steel Using Artificial Neural Network

Cold forged parts are mainly employed in automotive and aerospace assemblies, and strength plays an essential role in such applications. Backward extrusion is one such process in cold forging for the production of axisymmetrical cup-like parts, which is affected by a number of variables that influence the quality of the products. The study on the influencing parameters becomes necessary as the complexity of the part increases. The present paper focuses on the use of a multi-layered feed forward artificial neural network (ANN) model for determining the effects of process parameters such as billet size, reduction ratio, punch angle, and land height on forming behavior, namely, effective stress, strain, strain rate, and punch force in a cold forging backward extrusion process for AISI 1010 steel. Full factorial design (FFD) has been employed to plan the finite element (FE) simulations and accordingly, the input variables and response patterns are obtained for training from these FE simulations. This ANN model-based analysis reveals that the forming behavior of the cold forging backward extrusion process tends to increase with the billet size as well as the reduction ratios. However, decreases in punch angle and land height lead to the reduction of punch forces, which in turn enhances the punch life. FE simulation along with the developed ANN model scheme would benefit the cold forging industry in minimizing the process development effort in terms of cost and time.

Extending the Tsetlin Machine With Integer-Weighted Clauses for Increased Interpretability

Building models that are both interpretable and accurate is an unresolved challenge for many pattern recognition problems. In general, rule-based and linear models lack accuracy, while deep learning interpretability is based on rough approximations of the underlying inference. However, recently, the rule-based Tsetlin Machines (TMs) have obtained competitive performance in terms of accuracy, memory footprint, and inference speed on diverse benchmarks (image classification, regression, natural language understanding, and game-playing). TMs construct rules using human-interpretable conjunctive clauses in propositional logic. These, in turn, are combined linearly to solve complex pattern recognition tasks. This paper addresses the accuracy-interpretability challenge in machine learning by introducing a TM with integer weighted clauses - the Integer Weighted TM (IWTM). The intent is to increase TM interpretability by reducing the number of clauses required for competitive performance. The IWTM achieves this by weighting the clauses so that a single clause can replace multiple duplicates. Since each TM clause is formed adaptively by a Tsetlin Automata (TA) team, identifying effective weights becomes a challenging online learning problem. We solve this problem by extending each team of TA with another kind of automaton: the stochastic searching on the line (SSL) automaton. We evaluate the performance of the new scheme empirically using five datasets, along with a study of interpretability. On average, IWTM uses 6.5 times fewer literals than the vanilla TM and 120 times fewer literals than a TM with real-valued weights. Furthermore, in terms of average memory usage and F1-Score, IWTM outperforms simple Multi-Layered Artificial Neural Networks, Decision Trees, Support Vector Machines, K-Nearest Neighbor, Random Forest, Gradient Boosted Trees (XGBoost), Explainable Boosting Machines (EBMs), as well as the standard and real-value weighted TMs. IWTM finally outperforms Neural Additive Models on Fraud Detection and StructureBoost on CA-58 in terms of Area Under Curve, while performing competitively on COMPAS.

Measuring the Default Risk of Small Business Loans: Improved Credit Risk Prediction using Deep Learning

This paper suggests using a multilayer artificial neural network (ANN) method, known as deep learning ANN, to predict the probability of default (PD) within the survival analysis framework. Deep learning ANN structures consider hidden interconnections among the covariates determining the PD which can lead to prediction gains compared to parametric statistical methods. The application of the ANN method to a large data set of small business loans demonstrates prediction gains for the method relative to the logit and skewed logit models. These gains mainly concern short term prediction horizons and are more apparent for the type I misclassification error of loan default events, which has important implications for bank loans portfolio management. To identify the effects of covariates on the PD by the ANN structure, the paper proposes a bootstrap sampling method obtaining the distribution of changes of the PD over discrete covariate changes, while controlling for possible interactions among the covariates. We find that the covariates with the most important influence on the PD include the delinquent amount of a loan over its total balance, the payments and the balance of the loan over its installment, as well as the delinquency buckets of a loan. The duration of a loan is also found to be an important factor of default risk.