Chowdhury Mofizur Rahman

Hybrid decision tree and naïve Bayes classifiers for multi-class classification tasks

In this paper, we introduce two independent hybrid mining algorithms to improve the classification accuracy rates of decision tree (DT) and naive Bayes (NB) classifiers for the classification of multi-class problems. Both DT and NB classifiers are useful, efficient and commonly used for solving classification problems in data mining. Since the presence of noisy contradictory instances in the training set may cause the generated decision tree suffers from overfitting and its accuracy may decrease, in our first proposed hybrid DT algorithm, we employ a naive Bayes (NB) classifier to remove the noisy troublesome instances from the training set before the DT induction. Moreover, it is extremely computationally expensive for a NB classifier to compute class conditional independence for a dataset with high dimensional attributes. Thus, in the second proposed hybrid NB classifier, we employ a DT induction to select a comparatively more important subset of attributes for the production of naive assumption of class conditional independence. We tested the performances of the two proposed hybrid algorithms against those of the existing DT and NB classifiers respectively using the classification accuracy, precision, sensitivity-specificity analysis, and 10-fold cross validation on 10 real benchmark datasets from UCI (University of California, Irvine) machine learning repository. The experimental results indicate that the proposed methods have produced impressive results in the classification of real life challenging multi-class problems. They are also able to automatically extract the most valuable training datasets and identify the most effective attributes for the description of instances from noisy complex training databases with large dimensions of attributes.

An adaptive ensemble classifier for mining concept drifting data streams

It is challenging to use traditional data mining techniques to deal with real-time data stream classifications. Existing mining classifiers need to be updated frequently to adapt to the changes in data streams. To address this issue, in this paper we propose an adaptive ensemble approach for classification and novel class detection in concept drifting data streams. The proposed approach uses traditional mining classifiers and updates the ensemble model automatically so that it represents the most recent concepts in data streams. For novel class detection we consider the idea that data points belonging to the same class should be closer to each other and should be far apart from the data points belonging to other classes. If a data point is well separated from the existing data clusters, it is identified as a novel class instance. We tested the performance of this proposed stream classification model against that of existing mining algorithms using real benchmark datasets from UCI (University of California, Irvine) machine learning repository. The experimental results prove that our approach shows great flexibility and robustness in novel class detection in concept drifting and outperforms traditional classification models in challenging real-life data stream applications.

A complete OCR system for continuous Bengali characters

This paper is concerned with a complete optical character recognition (OCR) system for Bengali character. Recognition is done for both isolated and continuous printed multi font Bengali characters. Preprocessing steps includes segmentation in various levels, noise removal and scaling. Freeman chain code has been calculated from scaled character which is further processed to obtain a discriminating set of feature vectors for the recognizer. The unknown samples are classified using feed forward neural network based recognition scheme. It has been found from experimental results that success rate is approximately 98% for isolated characters and 96% for continuous character.

Adaptive Intrusion Detection based on Boosting and Naïve Bayesian Classifier

In this paper, we introduce a new learning algorithm for adaptive intrusion detection using boosting and naive Bayesian classifier, which considers a series of classifiers and combines the votes of each individual classifier for classifying an unknown or known example. The proposed algorithm generates the probability set for each round using naive Bayesian classifier and updates the weights of training examples based on the misclassification error rate that produced by the training examples in each round. This algorithm addresses the problem of classifying the large intrusion detection dataset, which improves the detection rates (DR) and reduces the false positives (FP) at acceptable level in intrusion detection. We tested the performance of the proposed algorithm with existing data mining algorithms by employing on the KDD99 benchmark intrusion detection dataset, and the experimental results proved that the proposed algorithm achieved high detection rates and significantly reduced the number of false positives for different types of network intrusions.

On the Power of Feature Analyzer for Signature Verification

This paper is concerned with verification of signatures using feature analysis and non linear classifier. Signatures are collected and scanned to obtain input image. Preprocessing involves removal of noise and making the input image size invariant. Feature analyzer can reduce the large domain of feature space and extract invariable information. Because of non linearity present in the input, a non linear classifier is used. Instead of using feed forward neural network, multiple feed forward neural networks are used which are trained in the form of ensemble. Using such ensemble makes the system more general than a regular single neural network based system. Resilient back propagation algorithm has been used for each neural network training to achieve faster recognition. Significant amount of training and testing has been performed using 10 fold cross validation and resultant impressive recognition accuracy (More than 90%) proves the effectiveness of the system.

Novel class detection in concept-drifting data stream mining employing decision tree

In this paper, we propose a new approach for detecting novel class in data stream mining using decision tree classifier that can determine whether an unseen or new instance belongs to a novel class. Most existing data mining classifiers can not detect and classify the novel class instances in real-time data stream mining problems like weather conditions, economical changes, astronomical, and intrusion detection etc, untill the classification models are trained with the labeled instances of the novel class. Arrival of a novel class in concept-drift occurs in data stream mining when new data introduce the new concept classes or remove the old ones. The proposed approach for incremental learning of concept drift considers mining, where the streaming data distributions change over time. It build a decision tree model from training dataset, which continuously updates so that the tree represents the most recent concept in data stream. The experiments on real benchmark data evaluate the efficiency of the proposed approach in both detecting the novel class and classification accuracy with comparisons of traditional data mining classifiers.

Prediction of State of Wireless Network Using Markov and Hidden Markov Model

Optimal resource allocation and higher quality of service is a much needed requirement in case of wireless networks. In order to improve the above factors, intelligent prediction of network behavior plays a very important role. Markov Model (MM) and Hidden Markov Model (HMM) are proven prediction techniques used in many fields. In this paper, we have used Markov and Hidden Markov prediction tools to predict the number of wireless devices that are connected to a specific Access Point (AP) at a specific instant of time. Prediction has been performed in two stages. In the first stage, we have found state sequence of wireless access points (AP) in a wireless network by observing the traffic load sequence in time. It is found that a particular choice of data may lead to 91% accuracy in predicting the real scenario. In the second stage, we have used Markov Model to find out the future state sequence of the previously found sequence from first stage. The prediction of next state of an AP performed by Markov Tool shows 88.71% accuracy. It is found that Markov Model can predict with an accuracy of 95.55% if initial transition matrix is calculated directly. We have also shown that O(1) Markov Model gives slightly better accuracy in prediction compared to O(2) MM for predicting far future.

Two-Level Dictionary-Based Text Compression Scheme

In this paper a new dictionary and memory based text compression technique is presented called a two-level dictionary based text compression scheme. The original words in a text file are transformed into codewords having length 2 and 3 using a dictionary comprising 73680 frequently used words in English language. Among these words most frequently used words use 2 length codewords and the rest use 3 length codewords for better compression. The codewords are chosen in such way that the spaces between words in the original text file can be removed altogether recovering a substantial amount of space. Another unique feature of our compression scheme is that we have recovered unused bit of ASCII character representation from each character to save one byte per 8 ASCII characters. Lastly a back end existing compression algorithm is used to finally compress the file. We have achieved about 75% (compression ratio of 2.01 bits per input character) reduction in size using our new compression strategy with gzip and bzip2.

Structure of Dictionary Entries of Bangla morphemes for morphological rule generation for Universal Networking Language

Dictionary plays a crucial role in any machine translation (MT) system. The Universal Networking Language (UNL) is an artificial language developed for conveying linguistic expressions in order to represent websites information into a standard form. In order to integrate Bangla into this platform it is necessary to develop both a dictionary and a grammar. This paper focuses on the development of a Structure of Dictionary Entries and Analysis of Grammatical Attributes of Bangla words such as Bangla Roots, Krit Prottoy (primary suffix) and Kria Bivokti (verb suffix). The goal is to make possible Bangla sentence encoversion to UNL and vice-versa. The theoretical analysis of our model proves that the proposed work is successfully able to prepare Universal words for Bangla roots, Krit Prottoy and Kria Bivokti along with their grammatical attributes for UNL.

Mining Complex Network Data for Adaptive Intrusion Detection

Intrusion detection is the method of identifying intrusions or misuses in a computer network, which compromise the confidentiality and integrity of the network. Intrusion Detection System (IDS) is a security tool used to monitor network traffic and detect unauthorized activities in the network [23, 28, 30]. A security monitoring surveillance system, which is an intrusion detectionmodel based on detecting anomalies in user behaviors was first introduced by James P. Anderson in 1980 [1]. After that several intrusion detection models based on statistics, Markov chains, time-series, etc proposed by Dorothy Denning in 1986. At first host-based IDSwas implemented, which located in the servermachine to examine the internal interfaces [35], but with the evolution of computer networks day by day focus gradually shifted toward network-based IDS [20]. Network-based IDS monitors and analyzes network traffics for detecting intrusions from internal and external intruders [26, 27, 34]. A number of data mining algorithms have been widely used by the intelligent computational researchers in the large amount of network audit data for detecting known and unknown intrusions in the last decade [3, 9, 18, 32, 33]. Even for a small network the amount of network audit data is very large that an IDS needs to examine. Use of data mining for intrusion detection aim to solve the problem of analyzing the large volumes of audit data and realizing performance optimization of detection rules.