AI-FML for Inference of Percentage of Votes Obtained and Real-life - - PowerPoint PPT Presentation
AI-FML for Inference of Percentage of Votes Obtained and Real-life Applications Team : Yangming and Dominic Presenters: Ying-Tai Chen, Seng-Yo Huang From: St. Dominic Catholic High School Yangming junior high school , Kaohsiung, Taiwan Outline
Team :Yangming and Dominic Presenters: Ying-Tai Chen, Seng-Yo Huang From: St. Dominic Catholic High School
Yangming junior high school , Kaohsiung, Taiwan
Introduction Machine learning Flowchart FML-based Knowledge Base & Rule Base Fuzzy Inference Mechanism Machine Learning – Genetic and PSO Algorithm Testing Conclusions
By using the FML intellectual decision tool to establish
Machine learning is executed by means of regression,
The goal of our project is to create a model used to predict the
final outcome of the election, an make those of the predicted results closer to the actual outcome of the election.
To compare the difference of the two algorithms, GA and
PSO, and perform testing based on the results of the elections.
Perform hard/software integration and utilize robotic
applications
Original data
Feature extraction Normalization
Training data
model
model
Testing data
Machine learning
Input
PA(Power Area) AD(Age Distribution) PS(Poll Satisfaction) PE(Political Experience)
Output
ET(Election)
Denotes the interdependence of the candidate’s strongholds and the electorate
since it enhances emotional identification
A bonus points system is used Three points is given when the candidate is born in the electorate Two points is given when the candidate is born in a different place, but the
candidate’s spouse is born in the same place, and
One point is given when there is no direct relativity between the candidate
and the electorate
According to findings of recent elections, young votes
dramatically influence the final outcome; whereas older people would have a rather fixed political preference
We used weighted mean to convert data to AD
20-29 yrs. Old multiplied by 5 30-39yrs. Old multiplied by 4 40-49yrs. Old multiplied by 3 50-59yrs. Old multiplied by 3 Beyond 60yrs. Old multiplied by 3
The Credits/ Discredits will directly affect the votes received
by the candidate
PS is calculated according to the average of the last
conducted poll survey
Denotes the political experience of the candidate since the
constituencies’ decisions are often based on the similar political experiences of the candidate
A bonus points system is used
Plus one point if worked as chief of village Plus two points if worked as member of parliament or Members of the
Legislative Yuan
Plus three points if worked as commissioner Plus four points if worked as deputy mayor Plus five points if worked as mayor of the city
The actual votes received by the candidate
PA、AD
Low:[0,0.1,0.3,0.4] Medium:[0.3,0.4,0.6,0.7] High:[0.6,0.7,0.9,1]
PS
Low:[0,0.1,0.2,0.3] Medium_Low:[0.2,0.3,.4,0.5] Medium_High:[0.4,0.5,0.7,0.8] High:[0.7,0.8,0.9,1]
PE
Low:[0,0.2,0.4,0.6] Medium_Low:[0.4,0.6,0.8,1] Medium_High:[0.8,1,1.4,1.6] High:[1.4,1.6,1.8,2]
ET
Low:[0,0.1,0.2,0.3] Medium_Low:[0.2,0.3,0.4,0.5] Medium:[0.4,0.5,0.6,0.7] Medium_High:[0.6,0.7,0.8,0.9] High:[0.8,0.9,1,1]
Selection:
We gathered data from numerous elections across Taiwan, and 66 of them became training data, and the rest of them became testing data(22).
PA/PE: Divide the whole column by 10. AD/PS/ET: Divide the whole column by 100 .
Orig. Aft.
Input Output
Using GA(Genetic Algorithm) Establish the knowledge base and input the rule base Using GA to perform machine learning(Training Data) and
Perform testing(Testing Data)
Generation
The correlation coefficient is a statistical measure of the strength of the
relationship between the relative movements of two variables. The values range between -1.0 and 1.0. A calculated number greater than 1.0 or less than - 1.0 means that there was an error in the correlation measurement. A correlation
perfect positive correlation. A correlation of 0.0 shows no linear relationship between the movement of the two variables.
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65
Inference Data
ET ET aft.
correl 0.842571
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65
Inference Data
ET ET aft.
correl 0.898741
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65
internal test
ET ET orig.
correl 0.850459
No PA AD PS PE ET ET orig. 1 0.130556 0.152 0.4 0.4586 2 0.3 0.475556 0.58 0.5 0.409 0.5414 3 0.3 0.057778 0.05 0.4 0.1095 4 0.3 0.510556 0.47 1 0.523 0.4948 5 0.3 0.172222 0.19 0.6 0.255 0.3823 6 0.3 0.237222 0.27 0.4 0.274 0.382 7 0.3 0.155 0.09 0.2768 8 0.3 0.292778 0.31 0.4 0.274 0.3229 9 0.3 0.161667 0.15 0.4 0.272 0.1798 10 0.3 0.356111 0.4 1 0.348 0.4118 11 0.1 0.02 0.01 0.0128 12 0.3 0.292222 0.41 0.2 0.274 0.3956 13 0.1 0.049444 0.046 0.5 0.0255 14 0.3 0.343333 0.31 0.8 0.266 0.3704 15 0.3 0.005 0.013 0.0054 16 0.3 0.001 0.04 0.0083 17 0.3 0.411667 0.4 0.6 0.517 0.5905 18 0.1 0.606111 0.59 0.6 0.515 0.5653 19 0.1 0.158889 0.17 0.2 0.274 0.2762 20 0.1 0.095556 0.07 0.0549 21 0.3 0.462222 0.44 1.1 0.527 0.6309 22 0.3 0.145556 0.15 0.8 0.245 0.3409 correl 0.846211
0.1 0.2 0.3 0.4 0.5 0.6 0.7 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
external test
ET ET orig.
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 1 2 3
2020 Election Testing
ET orig. ET aft.
ET orig. ET aft. 0.7003 0.7 0.0406 0.259 0.259 correl
0.99904716
Using PSO(Particle swarm optimization) Establish the knowledge base and input the rule base Using PSO to perform machine learning(Training Data) and
receive an optimized model
Perform testing(Testing Data)
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 1 2 3
PSO Algorithm
ET orig. ET AFT
ET orig. ET AFT . 0.7003 0.50059873 0.0406 0.09172902 0.259 0.32717 CORREL 0.96105583
Compare GA and PSO
A curvature design is shown in the KB diagram of GA, but
PSO doesn’t have this function up to now. And GA is also better since it shows a fuzzy concept during learning and inference
PSO has a colored design in the KB diagram which is
friendlier to new starters
After Fuzzy inference or fitness/accuracy calculations,
PSO directly indicates the corresponding level, therefore it’s easier to use speaking of this function.
According to the Internet, we could get reliable data
After comparing GA and PSO algorithm, we found that
Correlation is found within the two lists of testing
The isn’t always a relation between accuracy and
We found that there are still many variables that effects the
We hope that we can make FML accessible for everyone, and can
be used to predict what they wanted to know anywhere anytime.
What we did so far is just the mayor election, and we hope we
could predict presidential elections and even foreign elections.
We found that FML could be easily used and applied to everyday-
life, and we hope that we could develop a model that benefits our society.