Radwan E. Abdel-Aal

Short-term hourly load forecasting using abductive networks

Short-term load modeling and forecasting are essential for operating power utilities profitably and securely. Modern machine learning approaches, such as neural networks, have been used for this purpose. This paper proposes using the alternative technique of abductive networks, which offers the advantages of simplified and more automated model synthesis and analytical input-output models that automatically select influential inputs, provide better insight and explanations, and allow comparison with statistical and empirical models. Using hourly temperature and load data for five years, 24 dedicated models for forecasting next-day hourly loads have been developed. Evaluated on data for the sixth year, the models give an overall mean absolute percentage error (MAPE) of 2.67%. Next-hour models utilizing available load data up to the forecasting hour give a MAPE of 1.14%, outperforming neural network models for the same utility data. Two methods of accounting for the load growth trend achieve comparable performance. Effects of varying model complexity are investigated and proposals made for further improving forecasting performance.

GMDH-based feature ranking and selection for improved classification of medical data

Medical applications are often characterized by a large number of disease markers and a relatively small number of data records. We demonstrate that complete feature ranking followed by selection can lead to appreciable reductions in data dimensionality, with significant improvements in the implementation and performance of classifiers for medical diagnosis. We describe a novel approach for ranking all features according to their predictive quality using properties unique to learning algorithms based on the group method of data handling (GMDH). An abductive network training algorithm is repeatedly used to select groups of optimum predictors from the feature set at gradually increasing levels of model complexity specified by the user. Groups selected earlier are better predictors. The process is then repeated to rank features within individual groups. The resulting full feature ranking can be used to determine the optimum feature subset by starting at the top of the list and progressively including more features until the classification error rate on an out-of-sample evaluation set starts to increase due to overfitting. The approach is demonstrated on two medical diagnosis datasets (breast cancer and heart disease) and comparisons are made with other feature ranking and selection methods. Receiver operating characteristics (ROC) analysis is used to compare classifier performance. At default model complexity, dimensionality reduction of 22 and 54% could be achieved for the breast cancer and heart disease data, respectively, leading to improvements in the overall classification performance. For both datasets, considerable dimensionality reduction introduced no significant reduction in the area under the ROC curve. GMDH-based feature selection results have also proved effective with neural network classifiers.

Using GMDH-based networks for improved spam detection and email feature analysis

Unsolicited or spam email has recently become a major threat that can negatively impact the usability of electronic mail. Spam substantially wastes time and money for business users and network administrators, consumes network bandwidth and storage space, and slows down email servers. In addition, it provides a medium for distributing harmful code and/or offensive content. In this paper, we explore the application of the GMDH (Group Method of Data Handling) based inductive learning approach in detecting spam messages by automatically identifying content features that effectively distinguish spam from legitimate emails. We study the performance for various network model complexities using spambase, a publicly available benchmark dataset. Results reveal that classification accuracies of 91.7% can be achieved using only 10 out of the available 57 attributes, selected through abductive learning as the most effective feature subset (i.e. 82.5% data reduction). We also show how to improve classification performance using abductive network ensembles (committees) trained on different subsets of the training data. Comparison with other techniques such as neural networks and naive Bayesian classifiers shows that the GMDH-based learning approach can provide better spam detection accuracy with false-positive rates as low as 4.3% and yet requires shorter training time.

Construction and analysis of educational tests using abductive machine learning

Recent advances in educational technologies and the wide-spread use of computers in schools have fueled innovations in test construction and analysis. As the measurement accuracy of a test depends on the quality of the items it includes, item selection procedures play a central role in this process. Mathematical programming and the item response theory (IRT) are often used in automating this task. However, when the item bank is very large, the number of item combinations increases exponentially and item selection becomes more tedious. To alleviate the computational complexity, researchers have previously applied heuristic search and machine learning approaches, including neural networks, to solve similar problems. This paper proposes a novel approach that uses abductive network modeling to automatically identify the most-informative subset of test items that can be used to effectively assess the examinees without seriously degrading accuracy. Abductive machine learning automatically selects only effective model inputs and builds an optimal network model of polynomial functional nodes that minimizes a predicted squared error criterion. Using a training dataset of 1500 cases (examinees) and 45 test items, the proposed approach automatically selected only 12 items which classified an evaluation population of 500 cases with 91% accuracy. Performance is examined for various levels of model complexity and compared with that of statistical IRT-based techniques. Results indicate that the proposed approach significantly reduces the number of test items required while maintaining acceptable test quality.

Recognition of Handwritten Arabic (Indian) Numerals Using Freeman's Chain Codes and Abductive Network Classifiers

Accurate automatic recognition of handwritten Arabic numerals has several important applications, e.g. in banking transactions, automation of postal services, and other data entry related applications. A number of modelling and machine learning techniques have been used for handwritten Arabic numerals recognition, including Neural Network, Support Vector Machine, and Hidden Markov Models. This paper proposes the use of abductive networks to the problem. We studied the performance of abductive network architecture on a dataset of 21120 samples of handwritten 0-9 digits produced by 44 writers. We developed a new feature set using histograms of contour points chain codes. Recognition rates as high as 99.03% were achieved, which surpass the performance reported in the literature for other recognition techniques on the same data set. Moreover, the technique achieves a significant reduction in the number of features required.

Handwriting synthesis: classifications and techniques

Handwriting synthesis is the automatic generation of data that resemble natural handwriting. Although handwriting synthesis has recently gained increasing interest, the area still lacks a stand-alone review. This paper provides classifications for the different aspects of handwriting synthesis. It presents the applications, techniques, and evaluation methods for handwriting synthesis based on the several aspects that we identify. Then, it discusses various synthesis techniques. To the best of our knowledge, this paper is the only stand-alone survey on this topic, and we believe it can serve as a useful reference for the researchers in the field of handwriting synthesis.

Boosting paraphrase detection through textual similarity metrics with abductive networks

Graphical abstractDisplay Omitted HighlightsAnalyze a set of weak text reuse similarity metrics for paraphrase detection.Boost the performance of individual metrics using the abductive learning paradigm.Use decision-level fusion to build a committee of models of individual metrics.Use feature-level fusion to get a paraphrase detector using optimal set of metrics.Validate merits of the approach over individual metrics and other learning methods. A number of metrics have been proposed in the literature to measure text re-use between pairs of sentences or short passages. These individual metrics fail to reliably detect paraphrasing or semantic equivalence between sentences, due to the subjectivity and complexity of the task, even for human beings. This paper analyzes a set of five simple but weak lexical metrics for measuring textual similarity and presents a novel paraphrase detector with improved accuracy based on abductive machine learning. The objective here is 2-fold. First, the performance of each individual metric is boosted through the abductive learning paradigm. Second, we investigate the use of decision-level and feature-level information fusion via abductive networks to obtain a more reliable composite metric for additional performance enhancement. Several experiments were conducted using two benchmark corpora and the optimal abductive models were compared with other approaches. Results demonstrate that applying abductive learning has significantly improved the results of individual metrics and further improvement was achieved through fusion. Moreover, building simple models of polynomial functional elements that identify and integrate the smallest subset of relevant metrics yielded better results than those obtained from the support vector machine classifiers utilizing the same datasets and considered metrics. The results were also comparable to the best result reported in the literature even with larger number of more powerful features and/or using more computationally intensive techniques.

Abductive neural network modeling for hand recognition using geometric features

Hand recognition has received wide acceptance in many applications for automatic personal identification or verification in low to medium security systems. In this paper, we present a new approach for hand recognition based on abductive machine learning and hand geometric features. This approach is evaluated and compared to other learning algorithms including decision trees, support vector machines, and rule-based classifiers. Unlike other algorithms, the abductive learning approach builds simple polynomial neural network models by automatically selecting the most relevant features for each case. It also has acceptable accuracy with low false acceptance and false rejection rates. For the adopted dataset, the abductive learning approach has more than 98% overall accuracy, 1.67% average false rejection rate, and 0.088% average false acceptance rate.

Spam filtering with abductive networks

Spam messages pose a major threat to the usability of electronic mail. Spam wastes time and money for network users and administrators, consumes network bandwidth and storage space, and slows down email servers. In addition, it provides a medium to distribute harmful code and/or offensive content. In this paper, we investigate the application of abductive learning in filtering out spam messages. We study the performance for various network models on the spambase dataset. Results reveal that classification accuracies of 91.7% can be achieved using only 10 out of the available 57 content attributes. The attributes are selected automatically by the abductive learning algorithm as the most effective feature subset, thus achieving approximately 6:1 data reduction. Comparison with other techniques such as multi-layer perceptrons and naive Bayesian classifiers show that the abductive learning approach can provide better spam detection accuracies, e.g. false positive rates as low as 5.9% while requiring much shorter training times.

Comparison of Algorithmic and Machine Learning Approaches for the Automatic Fitting of Gaussian Peaks

Fitting Gaussian peaks to experimental data is important in many disciplines, including nuclear spectroscopy. Nonlinear least squares fitting methods have been in use for a long time, but these are iterative, computationally intensive, and require user intervention. Machine learning approaches automate and speed up the fitting procedure. However, for a single pure Gaussian, there exists a simple and automatic analytical approach based on linearisation followed by a weighted linear Least Squares (LS) fit. This paper compares this algorithmic method with an abductive machine learning approach based on AIM 1(Abductory Induction Mechanism). Both techniques are briefly described and their performance compared for analysing simulated and actual spectral peaks. Evaluated on 500 peaks with statistical uncertainties corresponding to a peak count of 100, average absolute errors for the peak height, position and width are 4.9%, 2.9% and 4.2% for AIM, versus 3.3%, 0.5% and 7.7% for the LS. AIM is better for the width, while LS is more accurate for the position. LS errors are more biased, under-estimating the peak position and over-estimating the peak width. Tentative CPU time comparison indicates a five-fold speed advantage for AIM, which also has a constant execution time, while LS time depends upon the peak width.