predictive analysis of health records using matlab

Post on 18-Jul-2015

271 Views

Category:

Data & Analytics

2 Downloads

Preview:

Click to see full reader

TRANSCRIPT

PREDICTIVE ANALYSIS OF HEALTH RECORDS

Umang ShuklaPranay SharmaKrishnan IyerMonica

➜ About dataset➜ What is fuzzy logic➜ Basics of fuzzy➜ Work done➜ Observations

OVERVIEW

ABOUT THE DATASET

Pima Indians Diabetes Database

Sources : UCI Machine Learning RepositoryOwners : National Institute of Diabetes and Digestive and Kidney DiseasesDonor : Vincent SigillitoDate Received : 9th May 1990Patients are females of age greater than 21 years of Pima Indian Heritage.Number of Instances: 768Number of Attributes: 8These attributes are :

1. Number of times pregnant 2. Plasma glucose concentration a 2 hours in an oral glucose tolerance test 3. Diastolic blood pressure (mm Hg) 4. Triceps skin fold thickness (mm) 5. 2-Hour serum insulin (mu U/ml) 6. Body mass index (weight in kg/(height in m)^2) 7. Diabetes pedigree function 8. Age (years) 9. Class variable (0 or 1)

What is Fuzzy Logic ?

Fuzzy logic is an approach to computing based on "degrees of truth" rather than the usual "true or false" boolean logic

It was first advanced by Dr. Lotfi Zadeh of the University of California

He said any logical system could be fuzzified.

LET’S REVIEW SOME CONCEPTS

Fuzzy Sets

Let X be a non empty set, A fuzzy set A in X is characterized by its membership function µA: X -> [0,1], where µA(x) is

the degree of membership of element x in fuzzy set A for each x ∈ X .

Operations

UnionIntersectionComplementComtainment

Membership Function

They map elements of a fuzzy set to real numbered values in the interval 0

to 1. Example:- Triangular, Trapezoidal, S-

shaped, Sigmoid, Pi-function

Fuzzification

The process of transforming crisp

(bivalued) input values into linguistic values is called

fuzzification

Defuzzification

Defuzzification converts the fuzzy values into crisp (bivalued) value.

Types :- Max-membership method Centroid method Weighted average method

WORKDONE

1.DATA CLEANING

2.SPLITTING OF

DATASET

3.BINNING OF

TRAINING SET

4.RULE GENERATION

USING J48

Classification tree by J48

5.FUZZIFICATION OF

BINNED INPUT AND OUTPUT

Fuzzy Inference System

6.DEFINING FUZZY

RULE

Fuzzified Rules

7.TEST SET

EVALUATION

EVALUATION TECHNIQUES

INTERACTIVE

This can be done by evalfis function on matlab

output= evalfis(input,fismat)

Evalfis() has the following arguments:➜ input: a number or a matrix specifying input values. ➜ fismat: an FIS structure to be evaluated.

ON MATLAB TERMINALEVALUATION TECHNIQUES

RESULTS AND OBSERVATION

Before understanding the results we need to know about the trapezoidal shaped member function which we used to define input variable.tramf = f(x,a,b,c,d)

OBSERVATIONS

In our test dataset we had 332 instances.

We evaluated our FIS model for 5%, 10%, 15% and 20% variance of the a, b, c, d point for each input member.

Next we took 0.65 as our membership value for output variable to classify predictions as “yes” or “no”.

However, It was observed that none of these changes in input variable boundary affected the accuracy of the predictions with exception to changes done in the member “plasma”.

On digging deeper we found out the reason for such a behaviour, we observed that even though accuracy was not changing these variance indeed affect the membership value of output but none of were big enough to cross the 6.5 barrier which we had set for output classification.

OBSERVATIONS

OBSERVATIONS

MODEL ACCURACY

J48 74.14%

DEFAULT(all with 5% variance) 80.722%

10% variance in plasma 81.024%

15% variance in plasma 81.626%

20% variance in plasma 81.626%

OBSERVATIONS

We observed that fuzzy system performs better than our J48 for same classification model as J48 uses crisp data values.

As only plasma was affecting the accuracy we found that it was so because of plasma was involved in all the rules defined above.

As we increased the input variance of plasma the accuracy showed an increase but only upto a particular level.

Thanks!Any questions?

top related