Вы находитесь на странице: 1из 14

Topics Covered:

Machine Learning
• BOD/COD predictions using GEP (Genetic Expression Programming Model)
• We attempted to apply the recently developed model GEP to use on our data.
• The current obtained results from GEP are presented in the further slides.
• I am still learning in depth about GEP and other possible models. We are
applying different models on our data so that we can compare their
performances and results.

Web Development
• Website Updated
• Database creation
• Data upload page

1
Our Problem lies in REGRESSION
Machine Learning

Supervised Learning Unsupervised Learning

Clustering
Classification

Regression

Classification Regression Clustering

2
Regression Models we applied so far:
Regression Models

Linear & Logistic Regression Srinivas Applied

Artificial Neural Network [Deep Learning] Srinivas Applied

Support Vector Regression Ritambhara Applied

Evolutionary Algorithms [Genetic Programming] Ritambhara Applied

3
GEP Model Applied on Our Data
We created two models:
• Model 1 (Organic Parameter):
pH, DO, Turbidity, Conductivity, Temperature
Narsapur Data Points: 29 Rajahmundry Data Points: 53

• Model 2 (Inorganic & Organic Parameters):


pH, DO, Turbidity, Conductivity, Temperature, Cl, Na, K, Ca, Mg, NO3
Narsapur Data Points: 16 Rajahmundry Data Points: 38

• Parameters Predicted:
COD, BOD

Assumptions:
Alternate rows are taken to make Training and Testing Data sets

4
Narsapur COD Predictions: Organic Parameters : pH, Temp, Conductivity,
Turbidity, DO

X-Axis: Predicted Value Y-Axis: Actual Value

• Train Data
R^2 = 0.84

• Test Data
R^2 = 0.19

5
Narsapur BOD Predictions: Organic Parameters : pH, Temp, Conductivity,
Turbidity, DO

X-Axis: Predicted Value Y-Axis: Actual Value

• Train Data
R^2 = 0.61

• Test Data
R^2 = 0.37

6
Rajahmundry COD Predictions: Organic Parameters : pH, Temp,
Conductivity, DO

X-Axis: Predicted Value Y-Axis: Actual Value

• Train Data
R^2 = 0.48
RMSE = 7.5

• Test Data
R^2 = 0.11
RMSE = 12.5

7
Rajahmundry BOD Predictions: Organic Parameters : pH, Temp, Conductivity,
Turbidity, DO

X-Axis: Predicted Value Y-Axis: Actual Value

• Train Data
R^2 = 0.59

• Test Data
R^2 = 0.05

8
Narsapur COD Predictions: Inorganic & Organic Parameters

pH, DO, Turbidity, Conductivity, Temperature, Cl, Na, K, Ca, Mg, NO3

X-Axis: Predicted Value Y-Axis: Actual Value

• Train Data
R^2 = 0.94

• Test Data
R^2 = 0.19

9
Narsapur BOD Predictions: Inorganic & Organic Parameters

pH, DO, Turbidity, Conductivity, Temperature, Cl, Na, K, Ca, Mg, NO3

X-Axis: Predicted Value Y-Axis: Actual Value

• Train Data
R^2 = 0.97

• Test Data
R^2 = 0.35

10
Rajahmundry COD Predictions: Inorganic & Organic Parameters

pH, DO, Turbidity, Conductivity, Temperature, Cl, Na, K, Ca, Mg, NO3

X-Axis: Predicted Value Y-Axis: Actual Value

• Train Data
R^2 = 0.73

• Test Data
R^2 = 0.23

11
Rajahmundry BOD Predictions: Inorganic & Organic Parameters

pH, DO, Turbidity, Conductivity, Temperature, Cl, Na, K, Ca, Mg, NO3

X-Axis: Predicted Value Y-Axis: Actual Value

• Train Data
R^2 = 0.72

• Test Data
R^2 = 0.06

12
Outcomes and Limitations
• The Models GEP predicted COD/BOD values have high errors and do
not show any noticeable trends yet.

Possible reasons of failures:


• Less number of data points (One missing cells can result in loss of one row
from the data.
• Data may not be perfectly accurate

• Future Plan?
• Getting more data
• Running algorithm with more permutations and combination of the current
parameters.
• Adding fluorescent sensor data to our models
• Comparison with different models (SVR, ANN) results.

13
Website Development
• Will is working on the 2D heat-maps on Thoreau Maps
• Ganga Data is updated on the Thoreau Maps

#TODO
• Apply Deep Learning Model on in-situ data and Lab Data
• Work with Will go get Databases up-to-date, and 2D
Thoreau Maps Platform running.

14

Вам также может понравиться