A. Rauf Baig

Opposition-Based Particle Swarm Optimization with Velocity Clamping (OVCPSO)

This paper presents an Opposition-based PSO(OVCPSO) which uses Velocity Clamping to accelerate its convergence speed and to avoid premature convergence of algorithm. Probabilistic opposition-based learning for particles has been used in the proposed method which uses velocity clamping to control the speed and direction of particles. Experiments have been performed upon various well known benchmark optimization problems and results have shown that OVCPSO can deal with difficult unimodal and multimodal optimization problems efficiently and effectively. The numbers of function calls (NFC) are significantly less than other PSO variants i.e. basic PSO with inertia weight, PSO with inertia weight and velocity clamping (VCPSO) and opposition based PSO with Cauchy Mutation (OPSOCM).

Measuring entertainment and automatic generation of entertaining games

Over the period of time computer games have became a major source of entertainment for humans. From the point of view of game developers there is a constant demand of writing games which are entertaining for the end users but entertainment itself is of subjective nature. It has always been difficult to quantify the entertainment value of the human player. The two factors which mainly influence the entertainment value are the type of the game and the contents of the game. In this paper we address the issues of measuring entertainment and automatic generation of computer games. We present some quantitative measures for entertainment in a genre of computer game and apply them as a guide for the evolution of new interesting games.

Honey bee foraging algorithm for multimodal & dynamic optimization problems

We present a new swarm based algorithm called Honey Bee Foraging (HBF). This algorithm is modeled after the food foraging behavior of the honey bees and performs a swarm based collective foraging for fitness in promising neighborhoods in combination with individual scouting searches in other areas. The strength of the algorithm lies in its continuous monitoring of the whole scouting and foraging process with dynamic relocation of the bees if more promising regions are found. The algorithm has the potential to be useful for optimization problems of multi-modal and dynamic nature.

Fraudulent call detection for mobile networks

Telecommunication industry has witnessed an enormous growth in terms of number of subscribers and revenue over the past few years. Still there are certain trends in the revenue of the telecommunication that show an instant fall, reason being change in customer behavior. Telecom operators are subjected to fraud in various forms, among the leading are subscription and superimposition fraud. In the U.S the sum of losses caused by fraudulent activity for the telecom industry is over 650 million dollars a year. The aim in this work is to cater the subscription fraud and bring the figures well within the desired range. In this work we use machine learning techniques to address the issue. Our solution uses a neural network to detect fraudulent behavior for subscription fraud. The neural network takes as input time series data of individual customers to predict their normal behavior. The crucial aspects of the networks predictions being accurate are the fraud profiles; some test cases are created which are used to make the neural network learn a fraudulent behavior.

Weighted MUSE for Frequent Sub-Graph Pattern Finding in Uncertain DBLP Data

Studies shows that finding frequent sub-graphs in uncertain graphs database is an NP complete problem. Finding the frequency at which these sub-graphs occur in uncertain graph database is also computationally expensive. This paper focus on investigation of mining frequent sub-graph patterns in DBLP uncertain graph data using an approximation based method. The frequent sub-graph pattern mining problem is formalized by using the expected support measure. Here n approximate mining algorithm based Weighted MUSE, is proposed to discover possible frequent sub-graph patterns from uncertain graph data

Evolutionary Search For Entertainment In Computer Games

Abstract Games have always been of interest to all age groups. With the advancement in technology and increase in number of users of personal computers, increased number of games is introduced in market. This is resulting in efforts, both for the developers in writing scripts for games and for the end users to select a game which is more entertaining. In this work we present a solution to both the issues. Initially a quantitative measure is devised, which calculates the entertainment value of games. Based upon the proposed measure we use evolutionary algorithm to generate games for different genres on the fly. The evolutionary algorithm needs to be given an initial set of games which it optimizes for entertainment using the proposed entertainment measure as the fitness criteria. In order to compare the entertainment value of the new games generated with the human’s entertainment value we conduct a human user survey.

Data Mining by Discrete PSO Using Natural Encoding

In this paper we have presented a new Discrete Particle Swarm Optimization approach to induce rules from the discrete data. Particles are encoded using Natural Encoding scheme. Encoding scheme and position update rule used by the algorithm allows individual terms corresponding to different attributes in the rule antecedent to be disjunction of values of those attributes. The performance of the proposed algorithm is evaluated against six different datasets using tenfold testing scheme. Achieved error rate has been compared against various evolutionary and non-evolutionary classification techniques. The algorithm produces promising results by creating highly accurate rules for each dataset.

A New Discrete PSO for Data Classification

In this paper we have presented a new Discrete Particle Swarm Optimization approach to induce rules from the discrete data. The proposed algorithm initializes its population by taking into account the discrete nature of the data. It assigns different fixed probabilities to current, local best and the global best positions. Based on these probabilities, each member of the population updates its position iteratively. The performance of the proposed algorithm is evaluated on five different datasets and compared against 9 different classification techniques. The algorithm produces promising results by creating highly accurate rules for each dataset.

Opposition Based Genetic Algorithm with Cauchy Mutation for Function Optimization

Evolutionary algorithms (EA) have been used in data classification and data clustering task since the advent of these algorithms. Nonlinear complex optimization problems have been the area of interest since very long time. The EA have been applied successfully on these optimization problems. The evolutionary algorithms suffer a lot due to their slow convergence rate, mainly due to evolutionary nature of these algorithms. This paper presents a new mutation scheme for opposition based genetic algorithms (OGA-CM). This scheme tunes the population during evolutionary process effectively by using Cauchy Mutation (CM). The performance of the algorithm is tested over suit of 5 functions. Opposition based Genetic Algorithm (OGA) is used as competitor algorithm to compare the results of the proposed algorithm. The results show that the proposed method outperforms GA and OGA for most of the test functions.

Performance Analysis of Frequent Itemset Mining Using Hybrid Database Representation Approach

Frequent itemset mining is considered to an important research oriented task in data mining, due to its large applicability in real world applications. In recent years lot of algorithms and techniques are proposed for enumerating itemsets from transactional databases. In which some are best for dense type datasets, while some are best for sparse type datasets. Currently there is no single algorithm exist that is best for all type of datasets (sparse as well as dense). The main limitation of previous algorithm is that, they depend upon single approach and do not combine the best features of multiple approaches for speedup the process of itemset mining. In this paper, the authors first compare and contract the two main itemset mining strategies on different itemset mining factors, scalability of algorithm, item search order, dataset projection and itemset frequency counting. Then the authors introduce a new hybrid strategy that combines the best features of existing strategies. Our different experiments on benchmark datasets show that mining all and maximal frequent itemsets using hybrid approach outperforms the previous algorithms on almost all types of dense and sparse datasets, which shows the effectiveness of our approach.