Load the data into a DataFrame

df $=$ pd$.$read$\_$csv$(\text{'}$seeds$.$csv$\text{'})$

Correlation Analysis

$$ Calculate the correlation values

$$ Reduce theNf x Nf $$f correlation DataFrame to non$-$redundant, non$-$identical feature pairs with the corresponding correlation value

$$ Examine and show the features pairings with a correlation value greater than $0\mathrm{.}7$

Partition Data

$$ Extract the features into a new DataFrame named X

$$ Extract the target labels into a DataFrame named y

$$ Make sure the target labels are in a DataFrame, not an array or Series

$$ Model predictions will be saved to this DataFrame

$$ Use double brackets: df$[[\text{'}$target$\text{'}]]$

Identify best k with elbow method

$$ Construct a function that produces the plot of SSE versus k The function should use a feature set X as input and return only the plot.

$$ Consider the following pseudo code:

from sklearn.cluster import KMeans

def calculate$\_$sse$\_$vs$\_$k$($X$)$:

## instantiate a list of array to hold the k ans SSE values at each iteration

sse$\_$v$\_$k $=[]$

## iterate over values of k

for k in np$.$arange$(1,21,1)$:

## instantiate and fit KMeans with the value of k and fixed random$\_$state

## add the k value and SSE $($inertia$\_)$ to the list

## plot the resulting sse versus k pairs

plt$.$figure$($figsize $=(9,5))$

plt$.$scatter$($x$=$ sse$\_$v$\_$k$[$:$,0],$ y$=$ sse$\_$v$\_$k$[$:$,1])$ ## show as points

plt$.$plot$($sse$\_$v$\_$k$[$:$,0],$sse$\_$v$\_$k$[$:$,1])$

plt$.$xlabel$(\text{'}$Cluster number $k$$\text{'})$

plt$.$ylabel$(\text{'}$SSE $($Inertia$)\text{'})$

plt$.$xticks$($ticks$=$ np$.$arange$(1,21,1))$

plt$.$show$()$

Show the SSE verus k plot for unscaled data

$$ What is the optimal value of k when clustering unscaled data?

Now, instantiate a StandardScaler, scale and transform x$,$ and show the SSE versus k for scaled data

$-$ What is the optimal value of k when clustering scaled data?

With optimal k values found, extract the cluster labels from KMeans Clustering

#### Unscaled data

$-$ Instantiate KMeans with the optimal k value found from the unscaled data. Be sure to employ the same random$\_$state that was used in the calculate$\_$sse$\_$vs$\_$k function above.

$-$ Fit this KMeans with the unscaled data

$-$ Extract the labels$\_$ from this KMeans and add these as a new column named $\text{'}$km$\_$label' to the target DataFrame

$-$ You may need to align the predicted labels to the actual labels

#### Scaled Data

$-$ Instantiate KMeans with the optimal k value found from the scaled data. Be sure to employ the same random$\_$state that was used in the calculate$\_$sse$\_$vs$\_$k function above

$-$ Fit this KMeans with the scaled data

$-$ Extract the labels$\_$ from this KMeans and add these as a new column named $\text{'}$km$\_$ss$\_$label' to the target DataFrame

$-$ You may need to align the predicted labels to the actual labels

With optimal k values found, extract the cluster labels from Agglomerative Clustering

#### Unscaled data

$-$ Instantiate AgglomerativeClustering with the optimal k value found from the unscaled data, and linkage$=$ 'complete'.

$-$ Fit this AgglomerativeClustering with the unscaled data

$-$ Extract the labels$\_$ from this AgglomerativeClustering and add these as a new column named 'agg$\_$label' to the target DataFrame

$-$ You may need to align the predicted labels to the actual labels

#### Scaled Data

$-$ Instantiate AgglomerativeClustering with the optimal k value found from the scaled data, and linkage$=$ 'complete'.

$-$ Fit this AgglomerativeClustering with the scaled data

$-$ Extract the labels$\_$ from this AgglomerativeClustering and add these as a new column named 'agg$\_$ss$\_$label' to the target DataFrame

$-$ You may need to align the predicted labels to the actual labels

## Compare the clustering results to k$-$Nearest Neighbors Classifier

$-$ Instantiate KNeighborsClassifier from sklearn.neighbors with default settings, fit to the unscaled data

$-$ Tip: The target dataset y has additional columns from clustering predictions. Be sure to call y$.$target

$-$ Add the predictions from the k$-$NN classifier and the unscaled data to the target DataFrame as knn$\_$label

$-$ Repeat for scaled data. Add those predictions as knn$\_$ss$\_$label

## Calculate the Accuracy Score and show the Confusion Matrix for all Predictors

$-$ There are six predictions to consider

$-$ KMeans clustering with unscaled and scaled data

$-$ Agglomerative clustering with unscaled and scaled data

$-$ k$-$NN Classifier with unscaled and scaled data

$-$ Which predictor performed the best?

$-$ Was there any case when the predictor with unscaled data outperformed the same predictor with scaled data?