Ctrl+K

演習・交通手段分類（機械学習）

9.6. 演習・交通手段分類（機械学習）#

9.6.1. 教師あり#

9.6.1.1. 分類#

交通手段をいくつかの特徴で分類します。

import geopandas as gpd
from sklearn.cluster import KMeans
from sklearn.decomposition import PCA
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score, confusion_matrix, classification_report
from sklearn.preprocessing import StandardScaler
from sklearn.tree import DecisionTreeClassifier
from sklearn import svm
from sklearn import tree
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt

# PLEASE REPLACE THE BELOW PATH WITH YOUR OWN
df = pd.read_csv('./traj_010_labeled_with_features.csv', index_col=0)
# create dataframe holding features per trip
# in this notebook, we focus on the features' mean scores per trip
df = df.groupby('trans_trip').agg({
    'distance': np.mean,
    'speed': np.mean,
    'accel': np.mean,
    'angle':np.mean,
    'angular_velocity':np.mean,
    'trans_mode':lambda x: pd.unique(x)[0],
})
df.head()

/var/folders/z5/0lnyp_m54dqc1xkz22ncbj2h0000gn/T/ipykernel_33103/2906762216.py:5: FutureWarning: The provided callable <function mean at 0x106508e00> is currently using SeriesGroupBy.mean. In a future version of pandas, the provided callable will be used directly. To keep current behavior pass the string "mean" instead.
  df = df.groupby('trans_trip').agg({
distance speed accel angle angular_velocity trans_mode
trans_trip
1.0 0.920653 55.720618 -0.014007 68.982383 0.354393 train
2.0 1.082321 65.670647 -0.001027 116.581851 0.568180 train
3.0 1.339320 74.683147 -0.003409 215.717754 0.480407 train
4.0 3.513432 173.776717 0.233277 174.470829 0.283810 train
5.0 4.282399 68.448484 -0.025508 150.209730 0.176290 train 
# drop rows holding np.nan
df.dropna(inplace = True)

df.describe()
distance speed accel angle angular_velocity
count 428.000000 428.000000 428.000000 428.000000 428.000000
mean 0.081362 39.136603 -0.068544 171.572185 24.004009
std 0.329114 38.591674 0.115817 64.819410 24.533810
min 0.001203 1.925925 -1.116834 5.303116 0.176290
25% 0.002057 6.276831 -0.108822 128.523012 2.852289
50% 0.009352 27.671732 -0.045742 173.816135 18.066445
75% 0.020917 53.299781 -0.006693 218.588914 35.412731
max 4.282399 175.182973 0.298885 329.530934 114.471672 
# Let's check the counts of the target categorical values in the dataset
df.trans_mode.value_counts()

trans_mode
walk      152
train      99
taxi       96
subway     47
bus        34
Name: count, dtype: int64

In this notebook, we merge several labels and use three labels: vehicle(car/taxi/bus), walk, and train(train/subway).

# create a dictionary holding new labels as values and corresponding old labels as keys
trans_mode_map = {'bus':'vehicle', 'car':'vehicle','taxi':'vehicle','subway':'train', 'walk':'walk','train':'train'}
# map the above dictionary to the current labels and replace them with the three labels
df['trans_mode'] = df.trans_mode.map(trans_mode_map)
# count each labels
df.trans_mode.value_counts()

trans_mode
walk       152
train      146
vehicle    130
Name: count, dtype: int64

### implementation of some basic classificaition models

create X holding features to be used in models and y holding labels to be predicted.

X = df[['speed','accel','angular_velocity']]
y = df['trans_mode']

# check correlations between features
X.corr()
speed accel angular_velocity
speed 1.000000 0.159520 -0.530622
accel 0.159520 1.000000 -0.068022
angular_velocity -0.530622 -0.068022 1.000000 
# check the sizes of X and y
X.shape, y.shape

((428, 3), (428,))

# Split into training and validation data.
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=1, stratify=y)

Run several classification models to estimate transportation mode (vehicle, train, and walk)

In this notebook, we run logistic regression, SVM(support vector machines), and Decision Tree as examples. To understand how each model works, please check the documents below and other machine learning introductory materials.

logistic regression: https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression

SVM: https://scikit-learn.org/stable/modules/svm.html#support-vector-machines

Decision Tree: https://scikit-learn.org/stable/modules/tree.html#decision-trees

# Logistic regression models (one-vs-rest)
lg_model=LogisticRegression()
lg_model.fit(X_train, y_train) # Fitting models to training data
lg_y_predicted=lg_model.predict(X_test) # Predicting labels with test data.
# Assessment of prediction accuracy
print('accuracy score (w/ training data): {}'.format(lg_model.score(X_train, y_train)))
# Assessment of prediction accuracy
print('accuracy score (w/ validation data): {}'.format(lg_model.score(X_test, y_test)))
print(classification_report(y_test, lg_y_predicted))
# print out confusion matrix
pd.DataFrame(confusion_matrix(y_test, lg_y_predicted), 
             columns = ['train','vehicle','walk'], 
             index=['train','vehicle','walk'])

accuracy score (w/ training data): 0.9130434782608695
accuracy score (w/ validation data): 0.9069767441860465
              precision    recall  f1-score   support

       train       0.92      0.80      0.85        44
     vehicle       0.80      0.92      0.86        39
        walk       1.00      1.00      1.00        46

    accuracy                           0.91       129
   macro avg       0.91      0.91      0.90       129
weighted avg       0.91      0.91      0.91       129

/opt/anaconda3/lib/python3.11/site-packages/sklearn/linear_model/_logistic.py:458: ConvergenceWarning: lbfgs failed to converge (status=1):
STOP: TOTAL NO. of ITERATIONS REACHED LIMIT.

Increase the number of iterations (max_iter) or scale the data as shown in:
    https://scikit-learn.org/stable/modules/preprocessing.html
Please also refer to the documentation for alternative solver options:
    https://scikit-learn.org/stable/modules/linear_model.html#logistic-regression
  n_iter_i = _check_optimize_result(
train vehicle walk
train 35 9 0
vehicle 3 36 0
walk 0 0 46

In the above cell, we printed out the accuracy score within training data and those within validation data.

Also, we printed out the classification reports that hold more information, including precision, recall, and f1-score.

  • \(\texttt{accuracy}(y, \hat{y}) = \frac{1}{n_\text{samples}} \sum_{i=0}^{n_\text{samples}-1} 1(\hat{y}_i = y_i)\)

  • \(\texttt{precision} = tp / (tp + fp)\) where tp is tp (true positive) is the correct result and fp (false positive) is an unexpected result

  • \(\texttt{recall} = tp / (tp + fn)\) where fn (false negative) is missing result

  • \(\texttt{f1-score} = 2 * (precision * recall) / (precision + recall)\)

With the confusion matrix, we can easily check which label are misclassified to which.

# Support Vector Machine (SVM)
svm_clf = svm.SVC()
svm_clf.fit(X_train, y_train) 
svm_y_predicted=svm_clf.predict(X_test) # Predicting labels with test data.
print("accuracy score (w/ training data): ", svm_clf.score(X_train, y_train)) 
print("accuracy score (w/ validation data): ", svm_clf.score(X_test, y_test)) 
print(classification_report(y_test, svm_y_predicted))
# print out confusion matrix
pd.DataFrame(confusion_matrix(y_test, svm_y_predicted), 
             columns = ['train','vehicle','walk'], 
             index=['train','vehicle','walk'])

accuracy score (w/ training data):  0.919732441471572
accuracy score (w/ validation data):  0.875968992248062
              precision    recall  f1-score   support

       train       0.92      0.82      0.87        44
     vehicle       0.79      0.79      0.79        39
        walk       0.90      1.00      0.95        46

    accuracy                           0.88       129
   macro avg       0.87      0.87      0.87       129
weighted avg       0.88      0.88      0.87       129
train vehicle walk
train 36 8 0
vehicle 3 31 5
walk 0 0 46 
# Decision Tree Classifier
tree_clf = DecisionTreeClassifier(random_state=0, max_depth=3)
tree_clf.fit(X_train, y_train) 
tree_y_predicted=tree_clf.predict(X_test)
print("accuracy score (w/ training data): ", tree_clf.score(X_train, y_train)) 
print("accuracy score (w/ validation data): ",tree_clf.score(X_test, y_test))
print(classification_report(y_test, tree_y_predicted))
# print out confusion matrix
pd.DataFrame(confusion_matrix(y_test, tree_y_predicted), 
             columns = ['train','vehicle','walk'], 
             index=['train','vehicle','walk'])

accuracy score (w/ training data):  0.9431438127090301
accuracy score (w/ validation data):  0.9302325581395349
              precision    recall  f1-score   support

       train       0.89      0.91      0.90        44
     vehicle       0.89      0.87      0.88        39
        walk       1.00      1.00      1.00        46

    accuracy                           0.93       129
   macro avg       0.93      0.93      0.93       129
weighted avg       0.93      0.93      0.93       129
train vehicle walk
train 40 4 0
vehicle 5 34 0
walk 0 0 46

Let’s visualize how the tree classifes our data

fig, ax = plt.subplots(figsize=(10, 5))
tree.plot_tree(
    tree_clf, ax = ax, fontsize=8, 
    feature_names= X.columns,  
    class_names= y.unique(), 
    filled=True,
    rounded=True,
)
plt.show()
../../_images/255af37f75b2d3e7a548f242b5818b92368ee05a33b29d4231932dc017731432.png

9.6.2. 教師なし#

ここではK-meansとPCAを用います

9.6.2.1. K-means#

ここでは、特徴量 angular_velocity と speed を用いて、K-means を用いてクラスタリングを行います。 K-meansでは、クラスタの数を指定する必要があるので、ここでは2つのクラスタを生成します。

X_kmeans = df[['angular_velocity', 'speed']]
model = KMeans(n_clusters=3) # with n_clusters specifying the number of clusters
model.fit(X_kmeans) # fitting the data
y_km=model.predict(X_kmeans) # predict clusters

/opt/anaconda3/lib/python3.11/site-packages/sklearn/cluster/_kmeans.py:870: FutureWarning: The default value of `n_init` will change from 10 to 'auto' in 1.4. Set the value of `n_init` explicitly to suppress the warning
  warnings.warn(

正解データとクラスタリングの結果を比較してみます

y = df['trans_mode']
y_colors = y.map({'train':'navy', 'vehicle':'turquoise', 'walk':'darkorange'}) 
fig, axs = plt.subplots(1, 2, figsize=(8, 3))
for i, ax in enumerate(axs):
    ax.scatter(df['angular_velocity'], df['speed'], c = [y_km, y_colors][i])
    ax.set_xlabel('angle rate')
    ax.set_ylabel('speed')
    ax.set_title(['K-means clustering', 'Ground truth class'][i])
plt.show()
../../_images/c24a3bc27d5862576602a2c3b26b258d486f2b73009ed8d9f35ef0b11c89c956.png

2つの特徴量しか用いていないこともあり、K-meanではあまりうまく分類ができていないですが、それでも大まかな傾向は捉えているようです。

9.6.2.1.1. PCA (Principal Component Analysis)#

高次元データの次元削減のために用いられることが多いです。 ここでは、GPS軌跡データの5つの特徴量をPCAを用いて2次元に縮小します。

X_pca = df[['distance','speed','accel','angle','angular_velocity']]
y = df['trans_mode']

X_pca = StandardScaler().fit_transform(X_pca)

pca = PCA(n_components=2)
X_pca_res = pca.fit_transform(X_pca)

print(pca.explained_variance_ratio_)

[0.36113444 0.2130784 ]

第1主成分と第2主成分をプロットします。

colors = ['navy', 'turquoise', 'darkorange']
target_names =  y.unique()
for color, target_name in zip(colors, target_names):
    plt.scatter(X_pca_res[y == target_name, 0], X_pca_res[y == target_name, 1], color=color, alpha=.8, label=target_name)
plt.legend(loc='best', shadow=False, scatterpoints=1, frameon=False)
plt.title('PCA of GPS dataset')
plt.show()
../../_images/c207a955827df77f077469063d2ac28b70240b432259a1a2cb2a71b8f920c44e.png
Contents