Setting up your environment<\/strong><\/h2>\nTo accomplish the solution presented in this article, we begin by setting up the correct environment in your machine to correctly execute the presented code. Here are some of the requirements that need to be configured in your environment:<\/p>\n
\n- \n
Windows 10<\/strong>. We are using Windows 10 as our main operating system.<\/p>\n<\/li>\n- \n
GridDB.<\/strong> We use GridDB as our main database that stores the data used in the machine learning model.<\/p>\n<\/li>\n- \n
Python 3.9.<\/strong> We use Python thanks Anaconda Navigator that allow deploying isolated working environments<\/p>\n<\/li>\n- \n
Jupyter Notebook<\/strong> is used as our main computing platform to execute Python cells.<\/p>\n<\/li>\n- \n
Microsoft Visual C++ 14.0<\/strong> or greater. This is needed for JPype1 used by GridDB to execute commands on Jupyter Notebook.<\/p>\n<\/li>\n- \n
Finally, we need to install any missing libraries that will be used in Jupyter Notebook<\/strong>. In our case, we were missing JPype1<\/code>, scikit-plot<\/code>, imblearn<\/code>. Here is how we install missing libraries, thanks to pip<\/code> command, in the Jupyter terminal:<\/p>\n<\/li>\n<\/ul>\n\n pip install imblearn \n pip install scikit-plot \n pip install JayDeBeApi JPype1==0.6.3 \n pip install pywaffle\n <\/code><\/pre>\n<\/div>\nOnce we have successfully installed and configured our environment, we can step right into our dataset.<\/p>\n
Introduction to the dataset<\/strong><\/h2>\nThe dataset used in this article contains 5110 records of patients. Each patient has 12 columns each referring to a concrete attribute. Most of these attributes correspond to medical records or the results of clinical trials. Some of the key attributes are hypertension, heart diseases, average glucose levels in the blood, and body mass index (BMI). As we can observe from these first attributes, the dataset provides relevant data regarding the likelihood of patients suffering from stroke disease. It\u2019s is easy to understand that a patient with high glucose levels and BMI, who has suffered from heart diseases and\/or hypertension, is more likely to suffer from stroke. In fact, stroke is also an attribute in the dataset and indicates in each medical record if the patient suffered from a stroke disease or not.<\/p>\n
The dataset is obtained from Kaggle <\/a>and is available for download. The following table provides an extract of the dataset used in this article.<\/p>\n![\"\"](\"https:\/\/github.com\/almahdibakkali96\/Stroke-Prediction-using-Machine-Learning-Python-and-GridDB-\/blob\/main\/stroke.PNG?raw=true\")
<\/p>\n
Importing the necessary libraries<\/strong><\/h2>\nIn this article we will be using multiple python libraries, that we will organize according to their usage:<\/p>\n
\n- Libraries for linear algebra and data processing: <\/li>\n<\/ul>\n\n
import numpy as np \n import pandas as pd\n <\/code><\/pre>\n<\/div>\n\n- Libraries for plotting graphs<\/li>\n<\/ul>\n\n
import matplotlib.pyplot as plt\n import matplotlib.ticker as mtick\n import matplotlib.gridspec as grid_spec\n import seaborn as sns\n from imblearn.over_sampling import SMOTE\n import scikitplot as skplt\n <\/code><\/pre>\n<\/div>\n\n- Libraries for building the model<\/li>\n<\/ul>\n\n
from sklearn.pipeline import Pipeline\n from sklearn.preprocessing import StandardScaler,LabelEncoder\n from sklearn.model_selection import train_test_split,cross_val_score\n \n from sklearn.metrics import classification_report, confusion_matrix\n from sklearn.metrics import accuracy_score, recall_score, roc_auc_score, precision_score, f1_score\n <\/code><\/pre>\n<\/div>\n\n- Libraries for evaluating the model using different algorithms<\/li>\n<\/ul>\n\n
from sklearn.linear_model import LinearRegression,LogisticRegression\n from sklearn.tree import DecisionTreeRegressor,DecisionTreeClassifier\n from sklearn.ensemble import RandomForestClassifier\n <\/code><\/pre>\n<\/div>\n\n- Library for reading data from GridDB in Jupyter Notebook<\/li>\n<\/ul>\n\n
import jaydebeapi\n <\/code><\/pre>\n<\/div>\nAfter we have successfully imported the required libraries, we begin by loading our dataset.<\/p>\n
Loading the Dataset<\/strong><\/h2>\nIn order to load the dataset we have two options that are described in more detail in the next paragraphs:<\/p>\n
For the purpose of this article, we have already designed and loaded our data into a GridDB container. The article about using Pandas with GridDB<\/a> can be useful if you want to write your data.<\/p>\nHowever, in this article, we do not read data from GridDB using Pandas, but we achieve this operation directly from the Jupyter Notebook, using the GridDB JDBC connector. The article on using GridDB in Jupyter<\/a> can be very useful if more details are needed on this step. Here is the code to achieve this:<\/p>\n\n def query_sensor(curs, table):\n curs.execute(\"select count(*) from \"+table)\n return curs.fetchall()[0][0]\n \n conn = jaydebeapi.connect(\"com.toshiba.mwcloud.gs.sql.Driver\", \"jdbc:gs:\/\/239.0.0.1:41999\/defaultCluster\", [\"admin\", \"admin\"], \"gridstore-jdbc-4.6.0.jar\")\n curs = conn.cursor()\n curs.execute(\"select * from \"#tables\"\")\n tables = []\n data = []\n \n for table in curs.fetchall():\n try:\n if table[1].split(\"_\")[1].startswith(\"M\"):\n tables.append(table[1])\n data.append(query_sensor(curs, table[1]))\n except:\n pass\n <\/code><\/pre>\n<\/div>\nYou can also load the dataset using the CSV file downloaded from Kaggle. Make sure to place the file in the root directory of your Jupyter project, and it should be ready to be used:<\/p>\n
\n df = pd.read_csv('healthcare-dataset-stroke-data.csv')\n <\/code><\/pre>\n<\/div>\nFor the purposes of this article, we will proceed with the data provided in the df<\/code> variable.<\/p>\nExploratory Data Analysis<\/strong><\/h2>\nBefore we proceed to build our machine learning model, we must begin with an exploratory data analysis that will allow us to find any inconsistencies in our data, as well as overall visualization of the dataset.<\/p>\n
First, we begin by checking for any null values in our dataset. This is achieved with the following lines of code:<\/p>\n
\ndf.isnull().sum()<\/code><\/pre>\n<\/div>\nThis cell outputs the following results, indicating that we have 201 missing values for the BMI attribute:<\/p>\n
\nid 0\ngender 0\nage 0\nhypertension 0\nheart_disease 0\never_married 0\nwork_type 0\nResidence_type 0\navg_glucose_level 0\nbmi 201\nsmoking_status 0\nstroke 0\ndtype: int64<\/code><\/pre>\n<\/div>\nIn this case, we can simply opt for removing the missing values, but we will use a decision tree in order to predict the BMI values for the missing entries, a method described in a Kaggle notebook<\/a>. This is achieved with the following code:<\/p>\n\nDT_bmi_pipe = Pipeline( steps=[ \n ('scale',StandardScaler()),\n ('lr',DecisionTreeRegressor(random_state=42))\n ])\nX = df[['age','gender','bmi']].copy()\nX.gender = X.gender.replace({'Male':0,'Female':1,'Other':-1}).astype(np.uint8)\n\nMissing = X[X.bmi.isna()]\nX = X[~X.bmi.isna()]\nY = X.pop('bmi')\nDT_bmi_pipe.fit(X,Y)\npredicted_bmi = pd.Series(DT_bmi_pipe.predict(Missing[['age','gender']]),index=Missing.index)\ndf.loc[Missing.index,'bmi'] = predicted_bmi<\/code><\/pre>\n<\/div>\nNow we have replaced all missing values, we can move to visualize our data. The first graph provides a numeric variable distribution of three key attributes that are age, average glucose levels, and BMI. This plot gives us a clear initial visualization of our data, being able to detect frequent age ranges, glucose levels, and BMIs. The code used to generate this plot is attached to the notebook in the public repository<\/a> for this article.<\/p>\n![\"\"](\"https:\/\/github.com\/almahdibakkali96\/Stroke-Prediction-using-Machine-Learning-Python-and-GridDB-\/blob\/main\/stroke1.png?raw=true\")
<\/p>\n
Windows 10<\/strong>. We are using Windows 10 as our main operating system.<\/p>\n<\/li>\n GridDB.<\/strong> We use GridDB as our main database that stores the data used in the machine learning model.<\/p>\n<\/li>\n Python 3.9.<\/strong> We use Python thanks Anaconda Navigator that allow deploying isolated working environments<\/p>\n<\/li>\n Jupyter Notebook<\/strong> is used as our main computing platform to execute Python cells.<\/p>\n<\/li>\n Microsoft Visual C++ 14.0<\/strong> or greater. This is needed for JPype1 used by GridDB to execute commands on Jupyter Notebook.<\/p>\n<\/li>\n Finally, we need to install any missing libraries that will be used in Jupyter Notebook<\/strong>. In our case, we were missing Once we have successfully installed and configured our environment, we can step right into our dataset.<\/p>\n The dataset used in this article contains 5110 records of patients. Each patient has 12 columns each referring to a concrete attribute. Most of these attributes correspond to medical records or the results of clinical trials. Some of the key attributes are hypertension, heart diseases, average glucose levels in the blood, and body mass index (BMI). As we can observe from these first attributes, the dataset provides relevant data regarding the likelihood of patients suffering from stroke disease. It\u2019s is easy to understand that a patient with high glucose levels and BMI, who has suffered from heart diseases and\/or hypertension, is more likely to suffer from stroke. In fact, stroke is also an attribute in the dataset and indicates in each medical record if the patient suffered from a stroke disease or not.<\/p>\n The dataset is obtained from Kaggle <\/a>and is available for download. The following table provides an extract of the dataset used in this article.<\/p>\n In this article we will be using multiple python libraries, that we will organize according to their usage:<\/p>\n After we have successfully imported the required libraries, we begin by loading our dataset.<\/p>\n In order to load the dataset we have two options that are described in more detail in the next paragraphs:<\/p>\n For the purpose of this article, we have already designed and loaded our data into a GridDB container. The article about using Pandas with GridDB<\/a> can be useful if you want to write your data.<\/p>\n However, in this article, we do not read data from GridDB using Pandas, but we achieve this operation directly from the Jupyter Notebook, using the GridDB JDBC connector. The article on using GridDB in Jupyter<\/a> can be very useful if more details are needed on this step. Here is the code to achieve this:<\/p>\n You can also load the dataset using the CSV file downloaded from Kaggle. Make sure to place the file in the root directory of your Jupyter project, and it should be ready to be used:<\/p>\n For the purposes of this article, we will proceed with the data provided in the Before we proceed to build our machine learning model, we must begin with an exploratory data analysis that will allow us to find any inconsistencies in our data, as well as overall visualization of the dataset.<\/p>\n First, we begin by checking for any null values in our dataset. This is achieved with the following lines of code:<\/p>\n This cell outputs the following results, indicating that we have 201 missing values for the BMI attribute:<\/p>\n In this case, we can simply opt for removing the missing values, but we will use a decision tree in order to predict the BMI values for the missing entries, a method described in a Kaggle notebook<\/a>. This is achieved with the following code:<\/p>\n Now we have replaced all missing values, we can move to visualize our data. The first graph provides a numeric variable distribution of three key attributes that are age, average glucose levels, and BMI. This plot gives us a clear initial visualization of our data, being able to detect frequent age ranges, glucose levels, and BMIs. The code used to generate this plot is attached to the notebook in the public repository<\/a> for this article.<\/p>\nJPype1<\/code>, scikit-plot<\/code>, imblearn<\/code>. Here is how we install missing libraries, thanks to pip<\/code> command, in the Jupyter terminal:<\/p>\n<\/li>\n<\/ul>\n pip install imblearn \n pip install scikit-plot \n pip install JayDeBeApi JPype1==0.6.3 \n pip install pywaffle\n <\/code><\/pre>\n<\/div>\nIntroduction to the dataset<\/strong><\/h2>\n
<\/p>\nImporting the necessary libraries<\/strong><\/h2>\n
\n
import numpy as np \n import pandas as pd\n <\/code><\/pre>\n<\/div>\n\n
import matplotlib.pyplot as plt\n import matplotlib.ticker as mtick\n import matplotlib.gridspec as grid_spec\n import seaborn as sns\n from imblearn.over_sampling import SMOTE\n import scikitplot as skplt\n <\/code><\/pre>\n<\/div>\n\n
from sklearn.pipeline import Pipeline\n from sklearn.preprocessing import StandardScaler,LabelEncoder\n from sklearn.model_selection import train_test_split,cross_val_score\n \n from sklearn.metrics import classification_report, confusion_matrix\n from sklearn.metrics import accuracy_score, recall_score, roc_auc_score, precision_score, f1_score\n <\/code><\/pre>\n<\/div>\n\n
from sklearn.linear_model import LinearRegression,LogisticRegression\n from sklearn.tree import DecisionTreeRegressor,DecisionTreeClassifier\n from sklearn.ensemble import RandomForestClassifier\n <\/code><\/pre>\n<\/div>\n\n
import jaydebeapi\n <\/code><\/pre>\n<\/div>\nLoading the Dataset<\/strong><\/h2>\n
def query_sensor(curs, table):\n curs.execute(\"select count(*) from \"+table)\n return curs.fetchall()[0][0]\n \n conn = jaydebeapi.connect(\"com.toshiba.mwcloud.gs.sql.Driver\", \"jdbc:gs:\/\/239.0.0.1:41999\/defaultCluster\", [\"admin\", \"admin\"], \"gridstore-jdbc-4.6.0.jar\")\n curs = conn.cursor()\n curs.execute(\"select * from \"#tables\"\")\n tables = []\n data = []\n \n for table in curs.fetchall():\n try:\n if table[1].split(\"_\")[1].startswith(\"M\"):\n tables.append(table[1])\n data.append(query_sensor(curs, table[1]))\n except:\n pass\n <\/code><\/pre>\n<\/div>\n df = pd.read_csv('healthcare-dataset-stroke-data.csv')\n <\/code><\/pre>\n<\/div>\ndf<\/code> variable.<\/p>\nExploratory Data Analysis<\/strong><\/h2>\n
df.isnull().sum()<\/code><\/pre>\n<\/div>\nid 0\ngender 0\nage 0\nhypertension 0\nheart_disease 0\never_married 0\nwork_type 0\nResidence_type 0\navg_glucose_level 0\nbmi 201\nsmoking_status 0\nstroke 0\ndtype: int64<\/code><\/pre>\n<\/div>\nDT_bmi_pipe = Pipeline( steps=[ \n ('scale',StandardScaler()),\n ('lr',DecisionTreeRegressor(random_state=42))\n ])\nX = df[['age','gender','bmi']].copy()\nX.gender = X.gender.replace({'Male':0,'Female':1,'Other':-1}).astype(np.uint8)\n\nMissing = X[X.bmi.isna()]\nX = X[~X.bmi.isna()]\nY = X.pop('bmi')\nDT_bmi_pipe.fit(X,Y)\npredicted_bmi = pd.Series(DT_bmi_pipe.predict(Missing[['age','gender']]),index=Missing.index)\ndf.loc[Missing.index,'bmi'] = predicted_bmi<\/code><\/pre>\n<\/div>\n<\/p>\n
![\"\"](\"https:\/\/github.com\/almahdibakkali96\/Stroke-Prediction-using-Machine-Learning-Python-and-GridDB-\/blob\/main\/stroke2.png?raw=true\")
<\/p>\n