3. INTRODUCTION
• Tumor prediction is a critical yet uphill task that requires proper
execution.
• Medical experts lack in-depth insight in this sector and at remote
places experts are not available to serve patients.
• Manually medical experts find it difficult to find the exact location and
proper medical solution to such tumors.
• Therefore these primary tumors detection and removal is very
critical.
• In healthcare sector machine learning plays an active role in
determining the disease risks and hence helps in diagnosis of
diseases.
• Hence machine learning can be of immense help to the clinical
expert team in developing intelligent and automated systems that
can predict the category of tumors at the earliest stage which can be
helpful in its treatment at the right time.
4. EVOLUTION OF BIOINSPIRED COMPUTING
APPROACH
• There are several problems that lack a complete precise solution.
• Even if the solution exists they are not cost effective and lack proper
analysis.
• Basically the issues with such problems are their solutions grow at a
rapid pace making it more complicated and hence feasible solutions
are hard to determine.
• In such scenarios Biologically Inspired Optimization Techniques may
be thought of as a more efficient tool to solve such optimization
problems.
• The root of such computation methodologies arise from nature by
evolution and by the activities of human brain to solve complicated
problems in engineering.
• Nature forms an immense source of motivation in determining
solution to sophisticated problem domain since nature dons the
features of the robustness, flexibility and dynamisms.
5. BIOINSPIRED COMPUTING DEFINED
• It frequently results in getting optimal solution. This is the basic idea
behind biologically motivated approach.
• These techniques may be thought of as a heuristic that exhibit the
characteristics of nature to solve a problem.
• These methods can be helpful in grafting solutions to problems that
contain uncertainty and noise where partial solutions exist.
• Bio-inspired approach minimizes the risk to get stuck in local
optimum by performing the search for solutions in various regions in
a parallel manner.
• Recently Biologically motivated techniques are emerging to be an
efficient computing method to solve imprecise and complex
problems.
• Such algorithm techniques can be applied to healthcare data for
effective diagnosis of tumor patients.
6. PROPOSED WORK
• Our research is based on implementation of biologically motivated
optimizing methods on tumor data to categorize tumor in to its
classes thereby helping the experts in Tumor prediction and
diagnosis.
• The primary tumor dataset that we have collected consists of 18
attributes and 339 instances.
• Our problem in hand is to accurately determine various 22 classes
of tumors.
• To achieve this we have used multilayer perceptron classifier. The
raw dataset may have many noisy features which may be irrelevant.
• Thus certain biologically inspired techniques like genetic search,
PSO search etc. have been applied to eliminate such features
before applying to classification process.
• The results have been evaluated using various performance metrics
7.
8.
9.
10.
11.
12.
13. IMPLEMENTATION ANALYSIS
• We have implemented biologically inspired computing to classify the
tumor data into various categories of tumor.
• Here we have formulated four different techniques to justify our
research.
• The nature inspired classification techniques include Multi-layer
Perceptron and the remaining three models are a hybrid
representation of Multi-layer Perceptron with different optimization
methods such as Genetic Search (GS), Particle Swarm Optimization
Search (PSO) and Evolutionary Search (ES).
• Performance parameters including Prediction Accuracy, RMSE,
Delay, Precision, Recall, F-Measure and Kappa Coefficient are
taken into consideration to evaluate efficiency of these Bio inspired
computing models.
14.
15.
16.
17.
18. CONCLUSION
• Bio inspired algorithms are one of the latest trend that can
revolutionize the world of computers.
• In this paper we have successfully implemented some critical bio-
inspired optimization methods and classified various categories of
primary tumor and our results were evaluated with some crucial
performance metrics which suggested that these techniques are a
positive force thereby guiding the medical experts in efficient
decision making.
19. An implementation of Feature
ranking using Machine learning
techniques for Diabetes disease
prediction
20. INTRODUCTION
• Diabetes disease is one of the crucial factors of death all over the
world.
• Machine learning methods are helpful in the diagnosis of diabetes
disease, showing a reasonable level of efficiency.
• But these data are redundant and are noisy in nature which
negatively affects the process of observing knowledge and useful
pattern.
• Further relevant data can be extracted from huge records using filter
based feature selection methods.
• In our study, a comparative analysis is drawn between four different
filter based feature selection methods.
• The main objective of our research is to make contributions in the
prediction of diabetes disease for healthcare research and analyze a
comprehensive comparison of popular filter based feature selection
methods.
21. FEATURE SELECTION
• Feature Selection methods are the optimizing agents in a machine
learning algorithm.
• Variable selection methods can be categorized into two parts:
Wrappers and Filters.
• The Wrapper selects and determines attributes based on accuracy
estimates by the target learning algorithm.
• While a filter method uses the statistical correlation between a set
of variable and the target variable.
22.
23.
24.
25.
26.
27. RESULTS AND DISCUSSION
• Our dataset comprises 9 attributes and 768 instances.
• It was subjected to four unique filter based feature selection
methods for the prediction of diabetes.
• Various parameters like Accuracy, TP Rate, FP Rate, Errors (RMSE,
MAE, RAE, and RRSE), Kappa Statistics, MCC and F-Score were
evaluated to determine the disease prediction efficiency.
28.
29.
30.
31. CONCLUSION
• In our research, a comparative detailed analysis was carried out on
the basis of filter based feature selection algorithms to predict the
risks of diabetes disease.
• Our study asserted that filter based attribute optimization methods
improve the performance of learning algorithms