Identifying Blood Donor Segments and Implementing Management Dashboard using Machine Learning and Data Visualization¶
For this project, we will be using machine learning specifically K-modes Clustering for blood donor segmentation - this jupyter notebook will show the step by step analysis and data wrangling done. Segmented data is then visualized using Power BI. Dataset used was from Taguig Pateros District Hospital
The study revolves around the data science question: How can we increase the number of donors in Blood Donation drives? Using the team's hypothesis By knowing the right people to invite, the following objective/solutions will be used as the direction of the study: 1) Blood Donor Segmentation, 2) Data Dashboard, and 3) Share EDA insights and propose initiatives.
This topic was chosen in the hopes that it can contribute to the still limited number of studies regarding Blood Donation initiatives in the Philippines. Moreover, the team knew that this analysis could be helpful for the Filipinos and that it can save lives.
Below are links to the supporting materials which will explain the study further:
- Capstone Project deck: https://drive.google.com/file/d/1AMgUk-4AX9geHX1uQ2VqV3YJhdGBd_30/view?usp=sharing
- Power BI recording: https://drive.google.com/file/d/1qKq6LkWHBEcbhxNR_oKc171IakppskFr/view?usp=sharing
Proponents of the project are Tine Celestial, Pamy Longos, Karen Salas, and Maico Rebong. This capstone project was created as a completion requirement for Analytiks Inc. bootcamp.
Import Libraries¶
In [1]:
# supress warnings
import warnings
warnings.filterwarnings('ignore')
# Importing all required packages
import numpy as np
import pandas as pd
# Data viz lib
import matplotlib.pyplot as plt
import seaborn as sns
%matplotlib inline
from matplotlib.pyplot import xticks
Load the Dataset¶
In [2]:
df = pd.read_csv(r"C:\Users\ACER\Desktop\Maico - Files\Data Science\Bootcamp Materials\Capstone\Blood_Donation_2019.csv", index_col = 'Donor ID')
Attribute Information(Categorical)
- Donor ID: Unique ID assigned to the donor
- Gender: Male or Female
- Blood Type: O+, A+, B+, AB+
- Barangay: 28 barangays in Taguig
- District: Congressional district, 1 or 2 only
- Network: Communications provider
- Quarter:Q1 to Q4 (full year 2019)
- Month: January to December (full year 2019)
- Day of the Week: Sunday to Saturday
- Donor Count: volume of blood donated by a donor, 450cc
- Blood Drive Location: Facility
- Data: TPDH
Exploratory Data Analysis (EDA)¶
In [3]:
# Check if the Data loaded correctly
df.head()
Out[3]:
In [4]:
df.tail()
Out[4]:
In [5]:
# Check the dimensions
df.shape
Out[5]:
In [6]:
# Check the features
df.info()
In [7]:
# Check for Column Names
df.columns
Out[7]:
In [8]:
# Get the general description of the data
df.describe(include='all')
Out[8]:
In [9]:
# Importing categorical columns
df_donor = df[['Gender', 'Blood Type', 'Barangay', 'District', 'Network', 'Month', 'Blood Drive Location']]
Data Inspection¶
In [10]:
df_donor.head()
Out[10]:
In [11]:
df_donor.tail()
Out[11]:
In [12]:
df_donor.shape
Out[12]:
In [13]:
df_donor.columns
Out[13]:
In [14]:
df_donor.info()
Data Cleaning¶
In [15]:
# Check the null values
df_donor.isnull().sum()
Out[15]:
In [16]:
# Drop null values
df_donor = df_donor.dropna()
df_donor = df_donor.drop(['District'], axis = 1)
In [17]:
df_donor.shape
Out[17]:
In [18]:
df_donor.info()
Model Building¶
In [19]:
# First we will keep a copy of data
df_donor_copy = df_donor.copy()
Data Preparation¶
In [20]:
from sklearn import preprocessing
le = preprocessing.LabelEncoder()
df_donor = df_donor.apply(le.fit_transform)
df_donor.head()
Out[20]:
Additional library¶
This project will be using k-modes as its partitioning algorithm since we are dealing with categorical data. Cao approach was used "to select initial cluster centers by considering the distance between objects and the density of each object."
In [21]:
from kmodes.kmodes import KModes
Using K-Mode with "Cao" initialization¶
In [22]:
km_cao = KModes(n_clusters=3, init = "Cao", n_init = 10, verbose=1)
fitClusters_cao = km_cao.fit_predict(df_donor)
In [23]:
# Predicted Clusters
fitClusters_cao
Out[23]:
In [24]:
clusterCentroidsDf = pd.DataFrame(km_cao.cluster_centroids_)
clusterCentroidsDf.columns = df_donor.columns
In [25]:
# Mode of the clusters
clusterCentroidsDf
Out[25]:
In [26]:
kmodes = km_cao.cluster_centroids_
shape = kmodes.shape
shape
Out[26]:
Choosing K by comparing Cost against each K¶
In [27]:
cost = []
for num_clusters in list(range(1,5)):
kmode = KModes(n_clusters=num_clusters, init = "cao", n_init = 10, verbose=1)
kmode.fit_predict(df_donor)
cost.append(kmode.cost_)
In [28]:
y = np.array([i for i in range(1,5,1)])
plt.plot(y,cost)
Out[28]:
Combining the predicted clusters with the original DF¶
In [29]:
df_donor = df_donor_copy.reset_index()
In [30]:
clustersDf = pd.DataFrame(fitClusters_cao)
clustersDf.columns = ['cluster_predicted']
combinedDf = pd.concat([df_donor, clustersDf], axis = 1).reset_index()
combinedDf = combinedDf.drop(['index'], axis = 1)
In [31]:
combinedDf.head()
Out[31]:
In [32]:
label = km_cao.labels_
In [33]:
combinedDf['cluster_name'] = ['cluster1' if x == 0 else\
'cluster2' if x == 1 else\
'cluster3' if x == 2 else\
'cluster4' for x in label]
In [34]:
plt.style.use('ggplot')
sns.countplot(combinedDf['cluster_name'],order=combinedDf['cluster_name'].value_counts().index,palette = 'rainbow')
plt.title('Cluster distribution')
plt.ylabel(None)
plt.xlabel(None)
Out[34]:
Cluster Identification¶
In [35]:
cluster_1 = combinedDf[combinedDf['cluster_predicted'] == 0]
cluster_2 = combinedDf[combinedDf['cluster_predicted'] == 1]
cluster_3 = combinedDf[combinedDf['cluster_predicted'] == 2]
In [36]:
cluster_1.info()
In [37]:
cluster_1.head()
Out[37]:
In [38]:
cluster_2.info()
In [39]:
cluster_2.head()
Out[39]:
In [40]:
cluster_3.info()
In [41]:
cluster_3.head()
Out[41]:
In [42]:
plt.style.use('ggplot')
plt.subplots(figsize = (15,5))
sns.countplot(x=combinedDf['cluster_name'],order=combinedDf['cluster_name'].value_counts().index,hue=combinedDf['Gender'])
plt.show()
In [43]:
plt.style.use('ggplot')
plt.subplots(figsize = (15,5))
sns.countplot(x=combinedDf['cluster_name'],order=combinedDf['cluster_name'].value_counts().index,hue=combinedDf['Blood Type'])
plt.show()
In [44]:
plt.style.use('ggplot')
plt.subplots(figsize = (15,12))
sns.countplot(x=combinedDf['cluster_name'],order=combinedDf['cluster_name'].value_counts().index,hue=combinedDf['Barangay'])
plt.show()
In [45]:
plt.style.use('ggplot')
plt.subplots(figsize = (15,5))
sns.countplot(x=combinedDf['cluster_name'],order=combinedDf['cluster_name'].value_counts().index,hue=combinedDf['Network'])
plt.show()
In [46]:
plt.style.use('ggplot')
plt.subplots(figsize = (15,10))
sns.countplot(x=combinedDf['cluster_name'],order=combinedDf['cluster_name'].value_counts().index,hue=combinedDf['Month'])
plt.show()
In [47]:
plt.style.use('ggplot')
plt.subplots(figsize = (15,10))
sns.countplot(x=combinedDf['cluster_name'],order=combinedDf['cluster_name'].value_counts().index,hue=combinedDf['Blood Drive Location'])
plt.show()
In [48]:
# Extract clustered data to csv
combinedDf.to_csv(r'C:\Users\ACER\Desktop\Maico - Files\Data Science\Bootcamp Materials\Capstone\Cao_v1_n3.csv')
Transform segmented data to Power BI visualization¶
Generate insights¶
Analyzing the segmented data, we were able to identify personas for each cluster. These are based on the dominant features (blood type, gender, barangay, donation facility, month, network) per grouping as well as the context of the donors.
- Cluster 1: Key group of new donors
- Cluster 2: Regular donors (Returnees)
- Cluster 3: New donors
Recommendations¶
Data-driven recommendations
- Gather more demographic information (ex. Sex, Age, Blood volume donated (cc), Employment Status) that can help improve segmentation.
- Conduct surveys to observe behavioral pattern of donors (ex. motivation to donate).
- Assign a health coordinator that would recruit blood donors local areas / barangays with lowest number of donors.
- Continue retention initiatives such as recognizing commitment in doing voluntary blood donation and partnering with companies for local areas / barnagay with highest number of donors.
- Increase in frequency of mobile blood drive during month of July and August. Target areas / barangays far from the in house facility.
Ways of working recommendations
- Automate follow-up invites for blood donations by creating a scheduling tool based on the last date of the donation + 3 months and the contact number provided in initial donation.
- Digitalize the donor data collection through use of registration laptop and online survey. To decentralize, you may opt to provide QR code directing to survey link and inspect “completed/submitted results” before proceeding.
- Advocacy talks Innovate materials to be used (ex. redesign slides, use of videos/testimonials), tie up with barangay for awareness sessions
Appendix¶
In [49]:
# Combined charts
for col in combinedDf:
plt.style.use('ggplot')
plt.subplots(figsize = (15,10))
sns.countplot(x='cluster_name', hue=col, data = combinedDf)
plt.show()
