Reference no: EM133212676
Assignment:
The PollutionNom.csv dataset provides age-adjusted mortality rate per 100,000 people for 60 locations. Additional climate and demographic information for each location is available as well.refer to the list below for attribute data types and summary attribute descriptions.
Population Data Set Descriptors:
Precipitation: JanuaryF: JulyF:
>65: Household: Education: Housing: Density: NonWhite: WhiteCollar: LowIncome: HC:
NOX: SO2: Humidity: Mortality:
Average annual precipitation in inches
Average January temperature in degrees F
Average July temperature in degrees F
% of population aged 65 or older
Average household size
Median school years completed by those over 22
% of housing units which are sound & with all facilities Population per sq. mile in urbanized areas
% non-white population in urbanized areas
% employed in white collar occupations
% of families with income < $3000
Relative hydrocarbon pollution potential
Relative nitric oxides pollution potential
Relative sulphur dioxide pollution potential
Annual average % relative humidity at 1pm
Total age-adjusted mortality rate per 100,000
Use Weka to answer the following questions. (Always use "Use training set" option for testing).
Clustering
1) Perform SimpleKMeans clustering with default parameters (2 clusters). How would you describe the two clusters based on the attribute characteristics? Interpret how the identified clusters are different based on average attribute values. Which attributes were more important to differentiate the clusters?
2) Perform SimpleKMeans clustering with three clusters. How would you describe the three clusters based on the attribute characteristics? Discuss which subsets of the population each cluster represents.
Neural Networks
1) Perform neural network analysis (MultilayerPerceptron) with two hidden layers ("hiddenLayers"=2). What is the overall prediction accuracy? Identify the attributes that significantly impact each of the two hidden nodes. How would you characterize these two hidden factors identified by the neural network analysis?
2) Repeat the same analysis with three hidden layers. What is the new prediction accuracy? Interpret the confusion matrix. Why do you think the accuracy is different? Identify the attributes that significantly impact each of the three hidden nodes. How would you characterize these three hidden factors identified by the neural network analysis?