Sina Khanmohammadi

An improved overlapping k-means clustering method for medical applications

The sensitivity of overlapping k-means algorithm to initialization is considered.The k-harmonic means method is effective for identifying initial cluster centroids.The proposed approach outperforms the original overlapping k-means algorithm. Data clustering has been proven to be an effective method for discovering structure in medical datasets. The majority of clustering algorithms produce exclusive clusters meaning that each sample can belong to one cluster only. However, most real-world medical datasets have inherently overlapping information, which could be best explained by overlapping clustering methods that allow one sample belong to more than one cluster. One of the simplest and most efficient overlapping clustering methods is known as overlapping k-means (OKM), which is an extension of the traditional k-means algorithm. Being an extension of the k-means algorithm, the OKM method also suffers from sensitivity to the initial cluster centroids. In this paper, we propose a hybrid method that combines k-harmonic means and overlapping k-means algorithms (KHM-OKM) to overcome this limitation. The main idea behind KHM-OKM method is to use the output of KHM method to initialize the cluster centers of OKM method. We have tested the proposed method using FBCubed metric, which has been shown to be the most effective measure to evaluate overlapping clustering algorithms regarding homogeneity, completeness, rag bag, and cluster size-quantity tradeoff. According to results from ten publicly available medical datasets, the KHM-OKM algorithm outperforms the original OKM algorithm and can be used as an efficient method for clustering medical datasets.

A Gaussian mixture model based discretization algorithm for associative classification of medical data

A new supervised discretization algorithm is proposed.Multi-modal distributed numerical variables/features are particularly handled.The proposed approach outperforms existing algorithms in rule-based classification. Knowledge-based systems such as expert systems are of particular interest in medical applications as extracted if-then rules can provide interpretable results. Various rule induction algorithms have been proposed to effectively extract knowledge from data, and they can be combined with classification methods to form rule-based classifiers. However, most of the rule-based classifiers can not directly handle numerical data such as blood pressure. A data preprocessing step called discretization is required to convert such numerical data into a categorical format. Existing discretization algorithms do not take into account the multimodal class densities of numerical variables in datasets, which may degrade the performance of rule-based classifiers. In this paper, a new Gaussian Mixture Model based Discretization Algorithm (GMBD) is proposed that preserve the most frequent patterns of the original dataset by taking into account the multimodal distribution of the numerical variables. The effectiveness of GMBD algorithm was verified using six publicly available medical datasets. According to the experimental results, the GMBD algorithm outperformed five other static discretization methods in terms of the number of generated rules and classification accuracy in the associative classification algorithm. Consequently, our proposed approach has a potential to enhance the performance of rule-based classifiers used in clinical expert systems.

A systems approach for scheduling aircraft landings in JFK airport

The aircraft landings scheduling problem at an airport has become very challenging due to the increase of air traffic. Traditionally, this problem has been widely studied by formulating it as an optimization model solved by various operation research approaches. However, these approaches are not able to capture the dynamic nature of the aircraft landing scheduling problem appropriately and handle uncertainty easily. A systems approach provides an alternative to solve such a problem from a systematic perspective. In this regard, the concept of general systems problem solving (GSPS) was first introduced in 1970s, and yet the power of the GSPS methodology is not fully discovered as it had only been applied to few domains. In this paper, a new general systems problem solving framework integrating computational intelligence techniques (GSPS-CI) is introduced. The two main functions of the framework are: (1) adaptive network based fuzzy inference system (ANFIS) to predict flight delays, and (2) fuzzy decision making procedure to schedule aircraft landings. The effectiveness of the GSPS-CI framework is tested on the JFK airport in USA, one of the most complex real-life systems.

AHP based Classification Algorithm Selection for Clinical Decision Support System Development

Abstract Supervised classification algorithms have become very popular because of their potential application in developing intelligent data analytic software. These algorithms are known to be sensitive to the characteristic and structure of input datasets, therefore, researchers use different algorithm selection methods to select the most suitable classification algorithm for specific dataset. These methods do not consider the uncertainty about input dataset, and relative importance of different performance measurements (such as speed, accuracy, and memory usage) in the target application domain. Therefore, these methods are not appropriate for software development. This is especially true in medical field where various high dimensional noisy data might be used with the software. Hence, software developers need to select one supervised classification algorithm that has the highest potential to provide good performance in wide variety of datasets. In this regard, an Analytic Hierarchy Process (AHP) based meta-learning algorithm is proposed to identify the most suitable supervised classification algorithm for developing clinical decision support system (CDSS). The results from ten publicly available medical datasets indicate that Support Vector Machine (SVM) has the highest potential to perform well on variety of medical datasets.

Prediction of Mortality and Survival of Patients After Cardiac Surgery Using Fuzzy EuroSCORE System and Reliability Analysis

Abstract Cardiac surgery is an important medical treatment for coronary vessel patients. Different models have been introduced to determine the risk factors related to side effects of this operation. The goal of this research is to study EuroSCORE (European System for Cardiac Operative Risk Evaluation) as a useful method for predicting the risk of mortality after cardiac surgery, and to introduce a new way of inference, called Fuzzy EuroSCORE. In addition, a systems reliability analysis will be used to calculate the survival possibility of patients after a certain time period after cardiac surgery. To model and simulate the suggested system, eight important parameters of EuroSCORE table are chosen using experts knowledge and a new method is applied based on a fuzzy inference system. To calculate the risk of mortality after cardiac surgery, the patients are categorized into 3 different groups of low risk, medium risk, and high risk. The range of the mortality risk is determined by appropriate medical data in the fuzzy EuroSCORE system. Additionally, a defect density function for the cardiovascular problem is suggested using the systems reliability analysis. Finally, the prospect of patients survival after a certain time period after cardiac surgery is predicted.

A Fuzzy Inference Model for Predicting Irregular Human Behaviour During Stressful Missions

Abstract In this paper a hybrid fuzzy inference and transfer function modeling is used to predict the irregular human behavior during hard and stressful tasks such as dangerous military missions. A set of affecting factors such as missioners experience, fatigue, sunshine intensity, hungriness, thirstiness, psychological characteristics, affright, etc. may be taken to account. In this regard a dynamic system model is used to predict the convolution of the timed effects of different factors on irregular behavior of personnel during the mission. This approach of predicting irregular behavior or erroneous decision making of staff have serious usages in aerospace, military, social and similar projects where a wrong decision can have catastrophic outcome such as attempting to suicide by a pilot or killing civilians by a soldier in stressful situations. The effect of such behavior and decisions may even cause the failure of the overall project or mission. For example, killing civilians by a soldier can result to the overall failure of human terrain missions where the main objective is gaining trust between the local civilian population.

A Simple Distance Based Seizure Onset Detection Algorithm Using Common Spatial Patterns

Existing seizure onset detection methods usually rely on a large number of extracted features regardless of computational efficiency, which reduces their applicability for real-time seizure detection. In this study, a simple distance based seizure onset detection algorithm is proposed to distinguish seizure and non-seizure EEG signals. The proposed framework first applies the common spatial patterns (CSP) method to enhance the signal-to-noise ratio and reduce the dimensionality of EEG signals, and then uses the autocorrelation of the averaged spatially filtered signal to classify incoming signals into a seizure or non-seizure state. The proposed approach was tested using CHB-MIT dataset that contains continuous scalp EEG recordings from 23 patients. The results showed \(\sim \)95.87 % sensitivity with an average latency of 2.98 s and 2.89 % false detection rate. More interestingly, the average process time required to classify each window (1–5 s of EEG signals) was 0.09 s. The outcome of this study has a high potential to improve the automatic seizure onset detection from EEG recordings and could be used as a basis for developing real-time monitoring systems for epileptic patients.

An improved synchronization likelihood method for quantifying neuronal synchrony

Indirect quantification of the synchronization between two dynamical systems from measured experimental data has gained much attention in recent years, especially in the computational neuroscience community where the exact model of the neuronal dynamics is unknown. In this regard, one of the most promising methods for quantifying the interrelationship between nonlinear non-stationary systems is known as Synchronization Likelihood (SL), which is based on the likelihood of the auto-recurrence of embedding vectors (similar patterns) in multiple dynamical systems. However, synchronization likelihood method uses the Euclidean distance to determine the similarity of two patterns, which is known to be sensitive to outliers. In this study, we propose a discrete synchronization likelihood (DSL) method to overcome this limitation by using the Manhattan distance in the discrete domain (l1 norm on discretized signals) to identify the auto-recurrence of embedding vectors. The proposed method was tested using unidirectional and bidirectional identical/non-identical coupled Hénon Maps, a Watts-Strogatz small-world network with nonlinearly coupled nodes based on Kuramoto model and the real-world ADHD-200 fMRI benchmark dataset. According to the results, the proposed method shows comparable and in some cases better performance than the conventional SL method, especially when the underlying highly connected coupled dynamical system goes through subtle changes in the bivariate case or sudden shifts in the multivariate case.

Classification Analysis of Chronological Age Using Brief Resting Electroencephalographic (EEG) Recordings

The present study aims to build a classification model that discriminates between chronological ages of subjects based on resting-state electroencephalography (EEG) data collected from a community sample of 269 children aged 7 to 11. Specifically, spectral power densities in four classical frequency bands: Delta (0.5–3 Hz), Theta (4–7 Hz), Alpha (8–12 Hz) and Beta (14–25 Hz) were extracted for each electrode as features, and fed to three classification algorithms including logistic regression (LR), support vector machine (SVM), and least absolute shrinkage and selection operator (Lasso). In addition, principal component analysis (PCA) was used to reduce the dimensions of the feature space. The results demonstrated that SVM and Lasso evidenced better performance (maximal accuracy = 80.68 ± 2.01% by SVM and 77.82 ± 2.11% by Lasso) when applied to original feature space, but LR yielded the best performance with PCA (80.72 ± 1.73%). The accuracy of binary classification exhibited a decreasing trend with diminishing chronological gaps between the groups.