This notebook has been designed for the nurse care activity recognition challenge competition with the the aim of providing the basic knowledge of Human Activity Recognition by accelerometer.

It has been made by Le Nhat Tan.

Download ipybn

click here

Exploring the data

import pandas as pd
import numpy as np
%matplotlib inline  

First, we have to load the data and label file. In this tutorial, we only use the train data of 1 user to reduce the time of the whole process.

#Load data
df=pd.read_csv('acc_user2.csv')
df_label=pd.read_csv('label/label_train.csv')

Let's check what information the data contains.

df.head(5)

	subject_id	datetime	x	y	z
0	2	2018-06-11T10:45:39.643+0900	-1.723	0.651	9.625
1	2	2018-06-01T10:59:22.466+0900	-2.260	0.996	9.471
2	2	2018-06-07T17:20:35.638+0900	1.685	1.800	9.489
3	2	2018-06-11T17:31:06.863+0900	-1.762	-2.068	9.436
4	2	2018-06-19T10:11:05.374+0900	2.834	-8.849	0.001

We can see that the data file contains 5 columns: subject_id, datetime, and 3 coordinates of the accelerometer data.

df_label.head(5)

	id	user_id	activity_type_id	activity_type	target_id	activity2user_id	start	finish
0	2098	38	26	休憩	2	19790	NaN	NaN
1	2852	5	26	休憩	5	21912	NaN	2018-03-15 03:41:24 +0900
2	5487	25	26	休憩	25	28937	NaN	2018-04-12 22:29:33 +0900
3	8872	26	26	休憩	26	37322	NaN	2018-04-26 13:52:11 +0900
4	9672	26	26	休憩	26	39599	NaN	2018-05-01 13:58:43 +0900

For label file, we have 8 columns: id (label id), user_id, activity_type_id, activity_type (name), target_id (patients), activity2user_id, start and finish timestamp of the activity.

Visualization the data

df.plot()

<matplotlib.axes._subplots.AxesSubplot at 0x7f083800c0d0>

Check the bar plot of labels

activity_map = {
1: "Vital", 
2: "Meal / medication", 
3: "Oral care", 
4: "Excretion", 
5: "Bathing / wiping",
6: "Treatment", 
7: "Morning gathering / exercises", 
8: "Rehabilitation / recreation", 
9: "Morning care", 
10: "Daytime user response", 
11: "Night care", 
12: "Nighttime user response", 
13: "Family / guest response", 
14: "Outing response", 
19: "Get up assistance", 
20: "Change dressing assistance", 
21: "Washing assistance", 
27: "Emergency response such as acciden",
15: "Linen exchange", 
16: "Cleaning",
23: "Preparation and checking of goods", 
24: "Organization of medications",
17: "Handwriting recording", 
18: "Delegating / meeting", 
22: "Doctor visit correspondence", 
25: "Family / doctor contact",
26: "Break", 
28: "Special remarks / notes"
}

df_label["activity_type_e"]=df_label["activity_type_id"].map(activity_map)
df_label['activity_type_e'].value_counts().plot.bar()

<matplotlib.axes._subplots.AxesSubplot at 0x2b4108f9460>

Pre-processing the data

The missing value and duplicated value rows are dropped. But we can also use several methods to handle missing value, it depends on your pipelines.

All the values are sorted by datetime

df = df.dropna()
df = df.sort_values('datetime')
df = df.drop_duplicates()

In the label file, nan and duplicated value rows are also dropped.

df_label = df_label.drop_duplicates('id')
df_label = df_label.dropna()

Now, we only get the label of user we utilize

df_label_user=df_label[df_label['user_id']==df['subject_id'].sample(1).values[0]]

We change all the timestamp data into the same data type

df['datetime']= pd.to_datetime(df['datetime'], format='%Y-%m-%dT%H:%M:%S.%f%z')
df_label_user['start'] = pd.to_datetime(df_label_user['start'], format='%Y-%m-%d %H:%M:%S %z')
df_label_user['finish'] = pd.to_datetime(df_label_user['finish'], format='%Y-%m-%d %H:%M:%S %z')

We can check how the data change after pre-processing

df.head(5)

	subject_id	datetime	x	y	z
1054607	2	2018-05-31 08:26:31.126000+09:00	-9.423	0.076	0.227
1894989	2	2018-05-31 08:26:31.326000+09:00	-9.462	0.000	0.227
1065508	2	2018-05-31 08:26:31.526000+09:00	-9.347	0.076	0.227
1066550	2	2018-05-31 08:26:31.533000+09:00	-9.347	0.076	0.227
5875832	2	2018-05-31 08:26:31.726000+09:00	-9.385	0.076	0.235

df_label_user.head(5)

	id	user_id	activity_type_id	activity_type	target_id	activity2user_id	start	finish
47	4375	2	26	休憩	2	26018	2018-04-05 08:38:38+09:00	2018-04-05 08:38:39+09:00
362	5960	2	6	処置	29	30128	2018-04-14 14:47:41+09:00	2018-04-14 14:47:42+09:00
4006	9704	2	4	排泄	81	39664	2018-05-01 17:03:26+09:00	2018-05-01 17:03:50+09:00
4361	11412	2	4	排泄	31	43065	2018-05-04 18:29:20+09:00	2018-05-04 18:29:54+09:00
6394	19144	2	4	排泄	36	58459	2018-05-17 18:00:50+09:00	2018-05-17 18:02:48+09:00

Segmentation

We try to reset the index value

df_label_user=df_label_user.reset_index()
df_label_user=df_label_user.drop(columns='index')

df_label_user.head(5)

Segment the data by the timestamp given by label file.

Every segment windows are extracted by the start and finish time of the activity in label file.

seg_list = []
seg_label_list = []

for i in range(len(df_label_user)):
    
    seg = df[(df["datetime"] >=df_label_user['start'][i]) & (df["datetime"] <= df_label_user['finish'][i])]
    seg_label = df_label_user["activity_type_id"][i]
    if (len(seg)!=0):
        seg_list.append(seg)
        seg_label_list.append(seg_label)

Features Extraction

In this tutorial, we extract 4 main features: STD, Average, Max, Min of 3 coordinates.

def get_features(x_data):
    #Set features list
    features = []
    #Set columns name list
    DFclist=list(x_data.columns)

    #Calculate features (STD, Average, Max, Min) for each data columns X Y Z 
    for k in DFclist:
        # std
        features.append(x_data[k].std(ddof=0))
        # avg
        features.append(np.average(x_data[k]))
        # max
        features.append(np.max(x_data[k]))
        # min
        features.append(np.min(x_data[k]))
    return features

features_list = []
for i in seg_list:
  #Drop 2 columns 'subject_id' and 'datetime', we only use 3 columns x, y, z
  i = i.drop(columns=['subject_id','datetime'])
  features_list.append(get_features(i))

Training

The Random Forest Model is utilized in this tutorial

from sklearn.ensemble import RandomForestClassifier         
model_ml = RandomForestClassifier(n_estimators=500,n_jobs=-1)

Divide data into train and test file to evaluate the results

from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(features_list, seg_label_list, test_size=0.3, random_state=42)

Train the model

model_ml.fit(X_train, y_train)

RandomForestClassifier(bootstrap=True, ccp_alpha=0.0, class_weight=None,
                       criterion='gini', max_depth=None, max_features='auto',
                       max_leaf_nodes=None, max_samples=None,
                       min_impurity_decrease=0.0, min_impurity_split=None,
                       min_samples_leaf=1, min_samples_split=2,
                       min_weight_fraction_leaf=0.0, n_estimators=500,
                       n_jobs=-1, oob_score=False, random_state=None, verbose=0,
                       warm_start=False)

Check the results

from sklearn.metrics import classification_report
from sklearn.metrics import confusion_matrix

y_predict = model_ml.predict(X_test)
print(classification_report(y_test,y_predict))
confusion_matrix(y_test, y_predict)

              precision    recall  f1-score   support

           1       0.00      0.00      0.00         0
           4       0.60      0.33      0.43         9
           7       1.00      0.50      0.67         2
          10       0.42      0.71      0.53         7
          13       0.00      0.00      0.00         1

    accuracy                           0.47        19
   macro avg       0.40      0.31      0.32        19
weighted avg       0.54      0.47      0.47        19

/usr/local/lib/python3.7/dist-packages/sklearn/metrics/_classification.py:1272: UndefinedMetricWarning: Precision and F-score are ill-defined and being set to 0.0 in labels with no predicted samples. Use `zero_division` parameter to control this behavior.
  _warn_prf(average, modifier, msg_start, len(result))
/usr/local/lib/python3.7/dist-packages/sklearn/metrics/_classification.py:1272: UndefinedMetricWarning: Recall and F-score are ill-defined and being set to 0.0 in labels with no true samples. Use `zero_division` parameter to control this behavior.
  _warn_prf(average, modifier, msg_start, len(result))

array([[0, 0, 0, 0, 0],
       [0, 3, 0, 6, 0],
       [1, 0, 1, 0, 0],
       [0, 2, 0, 5, 0],
       [0, 0, 0, 1, 0]])

Now, your turn! Let's analyze the challenge dataset!