Mushrooms are a part of funguses that typically grow above ground. Many mushrooms are edible, but some are extremely toxic to humans. Edible mushrooms are very nutritious. Mushrooms can also be used for medicinal purposes. Many varieties of mushrooms are grown for consumption, but some are foraged, but must be closely inspected in order to ensure they are edible. As mushrooms come in various shapes and sizes, it is important to not eat a wild mushroom unless absolutely certain it is not poisonous.

While a machine learning model is unlikely to be 100% reliable in classifying poisonous and non-poisonous mushrooms, there may still be value in being able to cluster similar types of mushrooms.

The data was taken from this dataset, and contains categorical features of various mushroom properties.

Resources:

Data Citation:

UCI Machine Learning. (2016, December 1). Mushroom classification. Mushroom Classification. https://www.kaggle.com/datasets/uciml/mushroom-classification

Import Libraries

# Libraries
import pandas as pd
import numpy as np

## Dates and Times
import pytz
from dateutil import parser
from datetime import timezone,datetime,timedelta
from dateutil.relativedelta import relativedelta
from datetime import date 

## String manipulation
import string
import re #regex
import json #json parsing
## Data Generation
import random

# Data Manipulation
import pandas as pd
import numpy as np
## Vector Math
from numpy.linalg import norm
## Math
import math

# Data Visualization
import matplotlib as mpl
import matplotlib.pyplot as plt

# Time
import time

# Sklearn Pipeline
from sklearn.pipeline import Pipeline
from sklearn.compose import ColumnTransformer
from sklearn.preprocessing import OneHotEncoder, StandardScaler
from sklearn.model_selection import train_test_split

# Models
from sklearn.cluster import KMeans, SpectralClustering, AgglomerativeClustering, Birch

# Metrics
from sklearn.metrics import accuracy_score, auc, confusion_matrix, f1_score
from sklearn.metrics import precision_score, recall_score
from sklearn.metrics import confusion_matrix, ConfusionMatrixDisplay

Loading and Cleaning Data

This data is particularly interesting because all features are categorical, including poisonous/safe column, which is the target of prediction.

Only column 11, which is ā€˜stalk-rootā€™ has missing values. We can use a impute a separate category (ā€˜mā€™) for these missing values.

Overall, there are 8,124 mushrooms in this dataset.

mushrooms = pd.read_csv('mushroom/mushroom_csv.csv')
mushrooms_class = mushrooms['class'] #edibility class
mushrooms = mushrooms.drop(columns = {'class'})
mushrooms['stalk-root'] = mushrooms['stalk-root'].fillna('m') #Fill NANs with 'm' for missing

len(mushrooms)

8124

Below is the printed list of categorical features in this dataset, along with each columnā€™s category list. We can see that there can be a large number of categories within each feature.

print('Columns: ', len(mushrooms.columns))
print(list(mushrooms.columns))

Columns:  22
['cap-shape', 'cap-surface', 'cap-color', 'bruises%3F', 'odor', 'gill-attachment', 'gill-spacing', 'gill-size', 'gill-color', 'stalk-shape', 'stalk-root', 'stalk-surface-above-ring', 'stalk-surface-below-ring', 'stalk-color-above-ring', 'stalk-color-below-ring', 'veil-type', 'veil-color', 'ring-number', 'ring-type', 'spore-print-color', 'population', 'habitat']

for i in mushrooms.columns:
    print(i, ": ", np.unique(mushrooms[i]))

cap-shape :  ['b' 'c' 'f' 'k' 's' 'x']
cap-surface :  ['f' 'g' 's' 'y']
cap-color :  ['b' 'c' 'e' 'g' 'n' 'p' 'r' 'u' 'w' 'y']
bruises%3F :  ['f' 't']
odor :  ['a' 'c' 'f' 'l' 'm' 'n' 'p' 's' 'y']
gill-attachment :  ['a' 'f']
gill-spacing :  ['c' 'w']
gill-size :  ['b' 'n']
gill-color :  ['b' 'e' 'g' 'h' 'k' 'n' 'o' 'p' 'r' 'u' 'w' 'y']
stalk-shape :  ['e' 't']
stalk-root :  ['b' 'c' 'e' 'm' 'r']
stalk-surface-above-ring :  ['f' 'k' 's' 'y']
stalk-surface-below-ring :  ['f' 'k' 's' 'y']
stalk-color-above-ring :  ['b' 'c' 'e' 'g' 'n' 'o' 'p' 'w' 'y']
stalk-color-below-ring :  ['b' 'c' 'e' 'g' 'n' 'o' 'p' 'w' 'y']
veil-type :  ['p']
veil-color :  ['n' 'o' 'w' 'y']
ring-number :  ['n' 'o' 't']
ring-type :  ['e' 'f' 'l' 'n' 'p']
spore-print-color :  ['b' 'h' 'k' 'n' 'o' 'r' 'u' 'w' 'y']
population :  ['a' 'c' 'n' 's' 'v' 'y']
habitat :  ['d' 'g' 'l' 'm' 'p' 'u' 'w']

Data Exploration

fig, ax = plt.subplots()
ax.bar(mushrooms['habitat'].value_counts().index, mushrooms['habitat'].value_counts().values)
ax.set_xlabel('habitat')
ax.set_ylabel('count')
ax.set_title('Habitat Distribution')

Text(0.5, 1.0, 'Habitat Distribution')

png

fig, ax = plt.subplots()
ax.bar(mushrooms['cap-color'].value_counts().index, mushrooms['cap-color'].value_counts().values)
ax.set_xlabel('cap-color')
ax.set_ylabel('count')
ax.set_title('Cap Color Distribution')

Text(0.5, 1.0, 'Cap Color Distribution')

png

fig, ax = plt.subplots()
ax.bar(mushrooms['odor'].value_counts().index, mushrooms['odor'].value_counts().values)
ax.set_xlabel('odor')
ax.set_ylabel('count')
ax.set_title('Odor Distribution')

Text(0.5, 1.0, 'Odor Distribution')

png

Looking at a select few dimensionā€™s distributions, it appears that certain categories in each dimension are more frequent than others. We will need to make sure that the train test split will split the data to ensure full coverage of all of the dimensional categories.

corr_matrix = pd.get_dummies(mushrooms).corr()
corr_threshold = 0.5
for i in range(len(corr_matrix)):
    for j in range(len(corr_matrix)):
        if i == j:
            continue
        elif corr_matrix.iloc[i,j] > corr_threshold:
            print(corr_matrix.columns[i], 'and', corr_matrix.columns[j], 'correlation: %.3f' %(corr_matrix.iloc[i,j]))
            

cap-color_r and ring-type_f correlation: 0.576
cap-color_u and ring-type_f correlation: 0.576
bruises%3F_f and stalk-surface-above-ring_k correlation: 0.541
bruises%3F_f and stalk-surface-below-ring_k correlation: 0.531
bruises%3F_f and ring-type_e correlation: 0.506
bruises%3F_t and stalk-surface-above-ring_s correlation: 0.562
bruises%3F_t and stalk-surface-below-ring_s correlation: 0.506
bruises%3F_t and ring-type_p correlation: 0.767
odor_a and stalk-root_c correlation: 0.515
odor_f and stalk-surface-above-ring_k correlation: 0.584
odor_f and stalk-surface-below-ring_k correlation: 0.600
odor_f and ring-type_l correlation: 0.724
odor_f and spore-print-color_h correlation: 0.800
odor_l and stalk-root_c correlation: 0.515
odor_m and stalk-color-above-ring_c correlation: 1.000
odor_m and stalk-color-below-ring_c correlation: 1.000
odor_m and ring-number_n correlation: 1.000
odor_m and ring-type_n correlation: 1.000
odor_s and gill-color_b correlation: 0.531
odor_y and gill-color_b correlation: 0.531
gill-attachment_a and gill-color_o correlation: 0.547
gill-attachment_a and gill-color_y correlation: 0.536
gill-attachment_a and stalk-color-above-ring_o correlation: 0.955
gill-attachment_a and stalk-color-below-ring_o correlation: 0.955
gill-attachment_a and veil-color_n correlation: 0.671
gill-attachment_a and veil-color_o correlation: 0.671
gill-attachment_f and veil-color_w correlation: 0.935
gill-spacing_w and stalk-root_e correlation: 0.570
gill-spacing_w and population_a correlation: 0.508
gill-spacing_w and habitat_g correlation: 0.538
gill-size_n and gill-color_b correlation: 0.777
gill-size_n and stalk-root_m correlation: 0.602
gill-size_n and ring-type_e correlation: 0.538
gill-size_n and spore-print-color_w correlation: 0.635
gill-size_n and population_v correlation: 0.506
gill-color_b and odor_s correlation: 0.531
gill-color_b and odor_y correlation: 0.531
gill-color_b and gill-size_n correlation: 0.777
gill-color_b and stalk-root_m correlation: 0.784
gill-color_b and ring-type_e correlation: 0.721
gill-color_b and spore-print-color_w correlation: 0.806
gill-color_b and population_v correlation: 0.523
gill-color_e and population_c correlation: 0.523
gill-color_e and habitat_w correlation: 0.703
gill-color_o and gill-attachment_a correlation: 0.547
gill-color_o and stalk-color-above-ring_o correlation: 0.573
gill-color_o and stalk-color-below-ring_o correlation: 0.573
gill-color_r and spore-print-color_r correlation: 0.576
gill-color_y and gill-attachment_a correlation: 0.536
stalk-root_b and spore-print-color_h correlation: 0.508
stalk-root_b and habitat_d correlation: 0.503
stalk-root_c and odor_a correlation: 0.515
stalk-root_c and odor_l correlation: 0.515
stalk-root_c and population_n correlation: 0.515
stalk-root_c and habitat_m correlation: 0.618
stalk-root_e and gill-spacing_w correlation: 0.570
stalk-root_e and population_a correlation: 0.557
stalk-root_m and gill-size_n correlation: 0.602
stalk-root_m and gill-color_b correlation: 0.784
stalk-root_m and ring-type_e correlation: 0.623
stalk-root_m and spore-print-color_w correlation: 0.887
stalk-root_r and stalk-surface-below-ring_y correlation: 0.817
stalk-surface-above-ring_k and bruises%3F_f correlation: 0.541
stalk-surface-above-ring_k and odor_f correlation: 0.584
stalk-surface-above-ring_k and stalk-surface-below-ring_k correlation: 0.677
stalk-surface-above-ring_k and ring-type_l correlation: 0.678
stalk-surface-above-ring_k and spore-print-color_h correlation: 0.554
stalk-surface-above-ring_s and bruises%3F_t correlation: 0.562
stalk-surface-above-ring_s and stalk-surface-below-ring_s correlation: 0.530
stalk-surface-above-ring_s and ring-type_p correlation: 0.582
stalk-surface-above-ring_y and stalk-color-above-ring_y correlation: 0.577
stalk-surface-above-ring_y and veil-color_y correlation: 0.577
stalk-surface-below-ring_k and bruises%3F_f correlation: 0.531
stalk-surface-below-ring_k and odor_f correlation: 0.600
stalk-surface-below-ring_k and stalk-surface-above-ring_k correlation: 0.677
stalk-surface-below-ring_k and ring-type_l correlation: 0.692
stalk-surface-below-ring_k and spore-print-color_h correlation: 0.568
stalk-surface-below-ring_s and bruises%3F_t correlation: 0.506
stalk-surface-below-ring_s and stalk-surface-above-ring_s correlation: 0.530
stalk-surface-below-ring_s and ring-type_p correlation: 0.535
stalk-surface-below-ring_y and stalk-root_r correlation: 0.817
stalk-color-above-ring_b and ring-type_l correlation: 0.544
stalk-color-above-ring_c and odor_m correlation: 1.000
stalk-color-above-ring_c and stalk-color-below-ring_c correlation: 1.000
stalk-color-above-ring_c and ring-number_n correlation: 1.000
stalk-color-above-ring_c and ring-type_n correlation: 1.000
stalk-color-above-ring_e and population_c correlation: 0.523
stalk-color-above-ring_e and habitat_w correlation: 0.703
stalk-color-above-ring_n and ring-type_l correlation: 0.531
stalk-color-above-ring_o and gill-attachment_a correlation: 0.955
stalk-color-above-ring_o and gill-color_o correlation: 0.573
stalk-color-above-ring_o and stalk-color-below-ring_o correlation: 1.000
stalk-color-above-ring_o and veil-color_n correlation: 0.703
stalk-color-above-ring_o and veil-color_o correlation: 0.703
stalk-color-above-ring_w and stalk-color-below-ring_w correlation: 0.551
stalk-color-above-ring_y and stalk-surface-above-ring_y correlation: 0.577
stalk-color-above-ring_y and stalk-color-below-ring_y correlation: 0.577
stalk-color-above-ring_y and veil-color_y correlation: 1.000
stalk-color-below-ring_b and ring-type_l correlation: 0.544
stalk-color-below-ring_c and odor_m correlation: 1.000
stalk-color-below-ring_c and stalk-color-above-ring_c correlation: 1.000
stalk-color-below-ring_c and ring-number_n correlation: 1.000
stalk-color-below-ring_c and ring-type_n correlation: 1.000
stalk-color-below-ring_e and population_c correlation: 0.523
stalk-color-below-ring_e and habitat_w correlation: 0.703
stalk-color-below-ring_o and gill-attachment_a correlation: 0.955
stalk-color-below-ring_o and gill-color_o correlation: 0.573
stalk-color-below-ring_o and stalk-color-above-ring_o correlation: 1.000
stalk-color-below-ring_o and veil-color_n correlation: 0.703
stalk-color-below-ring_o and veil-color_o correlation: 0.703
stalk-color-below-ring_w and stalk-color-above-ring_w correlation: 0.551
stalk-color-below-ring_y and stalk-color-above-ring_y correlation: 0.577
stalk-color-below-ring_y and veil-color_y correlation: 0.577
veil-color_n and gill-attachment_a correlation: 0.671
veil-color_n and stalk-color-above-ring_o correlation: 0.703
veil-color_n and stalk-color-below-ring_o correlation: 0.703
veil-color_o and gill-attachment_a correlation: 0.671
veil-color_o and stalk-color-above-ring_o correlation: 0.703
veil-color_o and stalk-color-below-ring_o correlation: 0.703
veil-color_w and gill-attachment_f correlation: 0.935
veil-color_y and stalk-surface-above-ring_y correlation: 0.577
veil-color_y and stalk-color-above-ring_y correlation: 1.000
veil-color_y and stalk-color-below-ring_y correlation: 0.577
ring-number_n and odor_m correlation: 1.000
ring-number_n and stalk-color-above-ring_c correlation: 1.000
ring-number_n and stalk-color-below-ring_c correlation: 1.000
ring-number_n and ring-type_n correlation: 1.000
ring-number_t and habitat_w correlation: 0.551
ring-type_e and bruises%3F_f correlation: 0.506
ring-type_e and gill-size_n correlation: 0.538
ring-type_e and gill-color_b correlation: 0.721
ring-type_e and stalk-root_m correlation: 0.623
ring-type_e and spore-print-color_w correlation: 0.679
ring-type_f and cap-color_r correlation: 0.576
ring-type_f and cap-color_u correlation: 0.576
ring-type_l and odor_f correlation: 0.724
ring-type_l and stalk-surface-above-ring_k correlation: 0.678
ring-type_l and stalk-surface-below-ring_k correlation: 0.692
ring-type_l and stalk-color-above-ring_b correlation: 0.544
ring-type_l and stalk-color-above-ring_n correlation: 0.531
ring-type_l and stalk-color-below-ring_b correlation: 0.544
ring-type_l and spore-print-color_h correlation: 0.869
ring-type_n and odor_m correlation: 1.000
ring-type_n and stalk-color-above-ring_c correlation: 1.000
ring-type_n and stalk-color-below-ring_c correlation: 1.000
ring-type_n and ring-number_n correlation: 1.000
ring-type_p and bruises%3F_t correlation: 0.767
ring-type_p and stalk-surface-above-ring_s correlation: 0.582
ring-type_p and stalk-surface-below-ring_s correlation: 0.535
spore-print-color_h and odor_f correlation: 0.800
spore-print-color_h and stalk-root_b correlation: 0.508
spore-print-color_h and stalk-surface-above-ring_k correlation: 0.554
spore-print-color_h and stalk-surface-below-ring_k correlation: 0.568
spore-print-color_h and ring-type_l correlation: 0.869
spore-print-color_r and gill-color_r correlation: 0.576
spore-print-color_w and gill-size_n correlation: 0.635
spore-print-color_w and gill-color_b correlation: 0.806
spore-print-color_w and stalk-root_m correlation: 0.887
spore-print-color_w and ring-type_e correlation: 0.679
population_a and gill-spacing_w correlation: 0.508
population_a and stalk-root_e correlation: 0.557
population_c and gill-color_e correlation: 0.523
population_c and stalk-color-above-ring_e correlation: 0.523
population_c and stalk-color-below-ring_e correlation: 0.523
population_c and habitat_w correlation: 0.744
population_n and stalk-root_c correlation: 0.515
population_v and gill-size_n correlation: 0.506
population_v and gill-color_b correlation: 0.523
habitat_d and stalk-root_b correlation: 0.503
habitat_g and gill-spacing_w correlation: 0.538
habitat_m and stalk-root_c correlation: 0.618
habitat_w and gill-color_e correlation: 0.703
habitat_w and stalk-color-above-ring_e correlation: 0.703
habitat_w and stalk-color-below-ring_e correlation: 0.703
habitat_w and ring-number_t correlation: 0.551
habitat_w and population_c correlation: 0.744

Looking at the correlations between dimensions, some do look to be highly correlated, but this is potentially due uneven distribution of mushrooms within each dimension, which would create potentially only one, or a few, mushroom sample with the combination of features.

Train Test Split

The train test split will be on a 20% of the entire data, chosen randomly. Not all models will use the train-test split, but it is important to do, especially since there will be further comparisons between supervised models.

train_x, test_x, train_y, test_y = train_test_split(mushrooms, mushrooms_class, \
                                                    test_size=0.2, random_state=1)

print('Train Length:', len(train_x))
print('Test Length:', len(test_x))

Train Length: 6499
Test Length: 1625

Build Pipeline

We will build a simple pipeline with a OneHotEncoder transformer for all columns of the data.

# Data Preprocessing
categorical_features = list(mushrooms.columns)
categorical_transformers = Pipeline(
    steps=[('encoder', OneHotEncoder())]
)

preprocessor = ColumnTransformer(
    transformers = [
        ('cat', categorical_transformers, categorical_features)
    ]
)

pipeline = Pipeline(
    steps=[('preprocessor', preprocessor)]
)

train_x_m = pipeline.fit_transform(train_x)
test_x_m = pipeline.transform(test_x)
mushrooms_m = pipeline.fit_transform(mushrooms)

Clustering

We can use a variety of clusters from the sklearn module to create two clusters. Unfortunately, this data is fairly sparse, so Non-Negative Matrix Factorization will not be used.

We will first create a function to test the two combinations of cluster labels and actual labels. The function will return the best label combo, the highest accuracy score, and the re-labeled model labels.

def best_labs(train_y, model_labels):
    max_acc = float('-inf')
    max_cats = None
    y_cats = list(np.unique(train_y))
    best_temp = None
    for i in [y_cats, sorted(y_cats, reverse=True)]:
        temp = model_labels.copy()
        temp = [i[0] if itm == 0 else i[1] for itm in temp]
        acc = accuracy_score(train_y, temp)
        if acc > max_acc:
            max_cats = i
            max_acc = acc
            best_temp = temp
    return max_cats, max_acc, temp

start = time.time()
kmeans_model = KMeans(n_clusters=2, n_init='auto', random_state=10) #Random State 10 leads to good results
kmeans_model.fit(train_x_m)
kmeans_train_labels = kmeans_model.labels_
kmeans_labs, kmeans_acc, kmeans_train_labels2 = best_labs(train_y, kmeans_train_labels)
print('K-Means Train Accuracy: %.3f' %(kmeans_acc))
kmeans_test_labels = kmeans_model.predict(test_x_m)
kmeans_test_labels = [kmeans_labs[0] if itm == 0 else kmeans_labs[1] for itm in kmeans_test_labels]
print('K-Means Test Accuracy: %.3f' %(accuracy_score(test_y, kmeans_test_labels)))
end = time.time()
print('K-Means Time: %.2f' %(end - start))
print()

start = time.time()
agg_model = AgglomerativeClustering(n_clusters=2)
agg_model.fit(mushrooms_m.toarray())
agg_train_labels = agg_model.labels_
agg_labs, agg_acc, agg_train_labels2 = best_labs(mushrooms_class, agg_train_labels)
print('Agglomerative Clustering Accuracy: %.3f' %(agg_acc))
end = time.time()
print('Agglomerative Time: %.2f' %(end - start))
print()

start = time.time()
birch_model = Birch(n_clusters=2)
birch_model.fit(train_x_m)
birch_train_labels = birch_model.labels_
birch_labs, birch_acc, birch_train_labels2 = best_labs(train_y, birch_train_labels)
print('Birch Train Accuracy: %.3f' %(birch_acc))
birch_test_labels = birch_model.predict(test_x_m)
birch_test_labels = [birch_labs[0] if itm == 0 else birch_labs[1] for itm in birch_test_labels]
print('Birch Test Accuracy: %.3f' %(accuracy_score(test_y, birch_test_labels)))
print('Birch Combined Accuracy: %.3f' %(birch_acc*0.8 + accuracy_score(test_y, birch_test_labels)*0.2))
end = time.time()
print('Birch Time: %.2f' %(end - start))

K-Means Train Accuracy: 0.892
K-Means Test Accuracy: 0.890
K-Means Time: 0.21

Agglomerative Clustering Accuracy: 0.890
Agglomerative Time: 3.60

Birch Train Accuracy: 0.892
Birch Test Accuracy: 0.885
Birch Combined Accuracy: 0.890
Birch Time: 3.27

cm = confusion_matrix(test_y, kmeans_test_labels)
cmd = ConfusionMatrixDisplay(cm, display_labels = kmeans_labs)
fig, ax = plt.subplots(figsize=(8,10))
ax.set_title('K-Means Confusion Matrix')
cmd.plot(ax=ax)

cm = confusion_matrix(mushrooms_class, agg_train_labels2)
cmd = ConfusionMatrixDisplay(cm, display_labels = agg_labs)
fig, ax = plt.subplots(figsize=(8,10))
ax.set_title('Agglomerative Confusion Matrix')
cmd.plot(ax=ax)

cm = confusion_matrix(test_y, birch_test_labels)
cmd = ConfusionMatrixDisplay(cm, display_labels = birch_labs)
fig, ax = plt.subplots(figsize=(8,10))
ax.set_title('Birch Confusion Matrix')
cmd.plot(ax=ax)

<sklearn.metrics._plot.confusion_matrix.ConfusionMatrixDisplay at 0x26dcf969dd0>

png

png

png

The confusion matrices for each clustering algorithm leads to some interesting results. Both true positives and true negatives are important results, but false negatives, or the case where the true label was ā€˜pā€™ for poisonous but the model classified it as ā€˜eā€™ for edible, is much worse than the false positive case (edible mushrooms classified as poisonous). The K-Means algorithm classified 7 poisonous mushrooms as edible, but both the Agglomerative and Birch clustering has zero cases.

From this, we can calculate an additional metric: the False Negative Rate. For the Agglomerative and Birch clustering, the false negative rate would be 0. For the K-Menas clustering, the false negative rate is $7/(813+7)=7/820=0.0085$, which is very low, but could potentially result in someone becoming very sick or even killed.

We can also look at the time taken for the model to cluster as a metric. Both the Agglomerative and Birch clustering took just over 3 seconds, while the K-Means clustering only took 1/5th of a second.

Comparison to Supervised Classification

We can quickly train various classification algorithms and compare accuracy and model fit time.

from sklearn.linear_model import LogisticRegression
from sklearn.svm import SVC
from sklearn.ensemble import RandomForestClassifier

start = time.time()
log_model = LogisticRegression()
log_model.fit(train_x_m, train_y)
log_predict = log_model.predict(test_x_m)
print('Logistic Regression Accuracy: %.3f' %(accuracy_score(test_y, log_predict)))
end = time.time()
print('Logistic Regression Time: %.2f' %(end - start))

start = time.time()
svc_model = SVC()
svc_model.fit(train_x_m, train_y)
svc_predict = svc_model.predict(test_x_m)
print('SVC Accuracy: %.3f' %(accuracy_score(test_y, svc_predict)))
end = time.time()
print('SVC Time: %.2f' %(end - start))

start = time.time()
rf_model = RandomForestClassifier()
rf_model.fit(train_x_m, train_y)
rf_predict = rf_model.predict(test_x_m)
print('Random Forest Accuracy: %.3f' %(accuracy_score(test_y, rf_predict)))
end = time.time()
print('Random Forest Time: %.2f' %(end - start))

Logistic Regression Accuracy: 0.998
Logistic Regression Time: 0.06
SVC Accuracy: 1.000
SVC Time: 0.74
Random Forest Accuracy: 1.000
Random Forest Time: 0.44

As we can see from the results, supervised classification algorithms classify the mushrooms at a higher accuracy and have much lower modeling times overall.

Conclusion

The key takeaway from this project is that if a supervised model can be created, it will likely perform better, both in terms of accuracy and compute time, than an unsupervised model. Interestingly though, with unsupervised models, while not as accurate as the supervised models, the unsupervised models predicted more false positives (edible mushrooms classified as poisonous) than false negatives (poisonous mushrooms classified as edible), which is the ā€˜betterā€™ incorrect result. Additionally, the unsupervised cluster results are still interesting, as even without ā€˜trainingā€™ enough difference was found between edible and poisonous mushrooms to group them solely based on their listed features.

Overall, this project was interesting to complete. Working with a dataset that was completely categorical was interesting to deal with. Future improvements could be made by potentially reducing the overall number of features, but was not necessary as the dataset was fairly small overall.