Prasun Chakrabarti

An Application of Classification Techniques on Breast Cancer Prognosis

Breast cancer is the most frequently diagnosed cancer in USA; furthermore breast cancer is the second most frequent cause of death for women in the United States as well as in Asia. In USA 40,600 deaths from breast cancer in 2009, 400 were men.[1] Several well established tools are currently used to screen for breast cancer including clinical breast exams, mammograms, and ultrasound. Supervised training is a technique in which a set of representative input output pairs is presented to the network. Through an iterative algorithm, the interval network weights are adjusted to decrease the difference between the network prediction and the true result for the training cases. The test has been performed on the breast cancer dataset using three classification techniques: Bayes learner, Decision Tree and Neural Net. The experiment concludes that Neural Net performance is better than the Decision Tree classification and Naive Bayes classification for early detection of breast cancer with better accuracy and precision.

Prediction of Breast Cancer Biopsy Outcomes - An Approach using Machine Leaning Perspectives

Breast cancer is the most frequently diagnosed cancer in USA. Furthermore breast cancer is the second major cause of death for women in USA. Several well established tools are currently used for screening for breast cancer including clinical breast exam, mammograms and ultrasound. Mammography is one of the most effective in terms of accuracy and cost. However the low positive predicted value (PPV) of breast cancer biopsies resulting from mammograms leads to 70% unnecessary biopsies with benign outcomes. In order to reduce the large number of surgical biopsies of breast, several CAD based system has been proposed in the last decades. Using these systems the radiologist gets an aid on their decision to perform breast biopsies. The dataset used is based on BIRADS findings. Prior work achieves good result with decision tree and neural network. The paper use AutoMLP, BP (back propagation) neural network and support vector machine (SVM) approach to predict the outcomes of mammogram with better result. Using SVM the false biopsies should significantly reduced to only 13%.

Analytical Approach on Indian Classical Raga Measures by Feature Extraction with EM and Naive Bayes

analysis is the main task in the musical information retrieval (MIR) systems. In this paper an analytical study based on these MIR techniques has been carried out to perform analysis of the Indian classical music and Indian ragas. The ragas are further classified into various thaats and their pitch class profiles and statistical measures. This paper demonstrates the strategy by which the various raga can be categorized using these statistical measures. The choices of algorithm used are the EM algorithm and the Naive bayes algorithm. Indian classical music is very popular because of the musical styles and the emotions it can reveal. Thus MIR (musical information retrieval) and its musical analysis is a very good choice for the researchers who have both knowledge of music and computer background. This paper includes the Matlab programming environment and toolbox for the effective result simulations. The EM and naive bayes algorithm have been utilized and the open source platform has been used for the rest of the work. Keywordsalgorithm, naive bayes, Indian classical music, music information retrieval, classification, clustering.

Having Centralized Monitoring as a Service in Cloud Computing: A Study of Performance Aspects

A Centralized Monitoring as a Service (CMaaS) is a desired and necessary feature to be included in cloud computing. One of the concerns in having CMaaS from both user and provider’s perspectives would be that of performance implications. Carrying out a performance analysis thus, becomes an important task before suggesting a MaaS solution. A straight forward performance study would be to find out whether the inclusion of monitoring processes affects the normal user request processing or not. The paper studies the affects by forming a simulation environment. The studies will also help datacenters in deciding whether to have dedicated VMs allocated for monitoring or to have monitoring processes share the VMs allocated for processing user requests.

Extending Access Management to maintain audit logs in cloud computing

Considering the most often talked about security risks in cloud computing, like, security and compliance, viability, lack of transparency, reliability and performance issues. Bringing strong auditability in cloud services can reduce these risks to a great extent. Also, auditing, both internally and externally is generally required and sometimes unavoidable looking into the present day competition in the business arena. Auditing in web based and cloud based usage environments focuses mainly on cost of a service which determines the overall expenditure of the user organization. However, the expenditure can be controlled by a collaborative approach between the provider company and the user organization by constantly monitoring the end user access and usage of subscribed cloud services. Though, many cloud providers will claim of having a robust auditable feature, the generic verifiability with sustainable long term recording of usage logs do not exist at all. Certain access management models can be perfectly extended to maintain audit logs for long terms. However, maintaining long term logs certainly has storage implications, especially with larger organizations. The storage implications need to be studied.

Business Planning in the Light of Neuro-fuzzy and Predictive Forecasting

In this paper we have pointed out gain sensing on forecast based techniques.We have cited an idea of neural based gain forecasting. Testing of sequence of gain pattern is also verifies using statsistical analysis of fuzzy value assignment. The paper also suggests realization of stable gain condition using K-Means clustering of data mining. A new concept of 3D based gain sensing has been pointed out. The paper also reveals what type of trend analysis can be observed for probabilistic gain prediction.

A Tree-Based Graph Coloring Algorithm Using Independent Set

This paper introduces a tree data structure-based graph coloring algorithm. Algorithm explores vertices in the tree form to finds maximal independent set, than these independent sets are colored with minimum colors. Proposed algorithm is tested on various DIMACS standard of graph instances. Algorithm is design to solve graph coloring problem for high degree graphs, i.e. the proposed algorithm is highly efficient for those graphs which has number of edges to number of vertices ratio is very high. Worst and best case time complexity of proposed algorithm is also discussed in this paper.

Performance Analysis and Error Evaluation Towards the Liver Cancer Diagnosis Using Lazy Classifiers for ILPD

This paper, entails the various Lazy classifiers such as IBKLG, LocalKnn algorithm, RseslibKnn algorithm used for diagnosis of the liver cancer. The results have been noted in terms of both performance and errors. The performance analyzed based on the accuracy, precision and recall and error evaluation are based on the Mean absolute error, Root mean squared error, Relative absolute error and Root relative squared error. The LocalKnn is best in terms of accuracy and recall while IBKLG indicates best precision.

Exploring Structure Oriented Feature Tag Weighting Algorithm for Web Documents Identification

There are various ways of web page classification but they take higher time to compute with lesser accuracy. Hence, there is a need to invent an efficient algorithm in order to reduce time and increase web page classification result. It is generally find that a few tags like title can contain the principle substance of text, and these patterns may have an impact on the adequacy of text classification. Although, the most widely recognized text weighting calculations, called term frequency inverse documents frequency (TF-IDF) doesn’t consider the structure of website pages. To take care of this issue, another feature tags weighting calculation is put in advanced. It thinks about the web page structure data like title, Meta tags, head etc. also content the useful information. In this proposed study first web site pages data are pre-processed and find text weight using TFIDF, after that using feature tag weighting calculation, frequent and important tags will find; then on the basis of text weight and tags weight, web document will classify.

Novel Work of Diagnosis of Liver Cancer Using Tree Classifier on Liver Cancer Dataset (BUPA Liver Disorder)

The classification plays a vital role towards diagnosis of liver cancer because still diagnosis of liver cancer is tedious job in early stages and late stage it is incurable. In this paper, BUPA liver disorder has been used and the Tree classifier used result is analyzed into WEKA Tool. LMT, J48, Random Forest, REP tree, Extra tree, Simple cart algorithms are have been utilized to investigate towards performance (accuracy, precision and recall) and error evaluation (Mean absolute error, Root mean squared error, Relative absolute error, Root relative squared error) performed.