Refreshment Course for Data Science and Business Intelligence Certification

Introduction

When we are asked to prove our credibility in a field, for example, as doctors we are required to show our licenses. In many professions, certification is one of the main ways to demonstrate credibility.

For instance, there are internationally recognized certifications such as AWS or GCP. Many countries or third-party institutions also provide their own certifications. In Indonesia, an official government body is responsible for this. In this post, we will discuss some of their certifications in data science and business intelligence, as well as the refresher materials available for preparation.

Data Science Certification

Each certification comes with predefined units that the test taker must complete in order to become certified. For example, in Data Science, the required units include:

• Data Collection

• Data Exploration

• Data Validation

• Data Definition

• Data Construction

• Data Labeling

• Model Building

• Model Evaluation

While I am not entirely sure whether these units fully reflect the best practices of a modern data science workflow, I believe they could be simplified and better aligned with current standards, such as:

• Data Collection

• Data Understanding

• Data Exploration

• Data Preprocessing

• Model Development

• Model Deployment

It is also worth noting that these units were most likely designed before the pre-LLM era. Today, with the advancement of generative AI, many of these steps could potentially be automated or significantly accelerated. :D

Data Collection

The first step is data collection. When taking a certification, the provider usually provides both the dataset and the goal for the test takers. However, there may be cases where you need to find and collect the data yourself. If so, what should you do?

First, you need to determine what kind of data you want to analyze. For example:

• Stock data, if you are role-playing as an investment analyst.

• Revenue data, if you are role-playing as a company analyst aiming to boost revenue.

• Customer churn data, if your goal is to prevent the company from losing customers, etc.

It is important to remember that every type of data comes with its own characteristics, which will influence how you approach analysis.

Second, after deciding on the type of data, you should identify platforms to find open datasets, such as Kaggle. Alternatively, if you have access to proprietary data and permission to use it, you can work with that as long as privacy and confidentiality are preserved.

For simplicity, we will use the Telco customer churn dataset, which can be found [1].

Data Understanding

After deciding on the dataset, it is essential to understand what the data is about. I usually find that reading the data’s description and metadata is very helpful. Here is what the provider states about the Telco dataset:

Predict behavior to retain customers. You can analyze all relevant customer data and develop focused customer retention programs. The raw data contains 7043 rows (customers) and 21 columns (features)." [IBM Sample Data Sets]

From this, we know that the dataset contains 7,043 customers with 21 properties. Taking a quick glance at the data, we can notice some properties that are likely of most interest, Churn. By this time, we may come up with several questions, such as:

• What proportion of customers have churned?

• What are the main reasons customers decide to churn?

• How does customer churn affect the company?

• How can we prevent more customers from churning?

• What strategies should we implement?

From my perspective, data science is essentially a workflow designed to answer these questions. This can be achieved by building predictive models to simulate customer behavior and ultimately create the desired outcome; in this case, retaining more customers.

Data Exploration

After understanding the data and defining the objectives, the next step is usually to conduct Exploratory Data Analysis (EDA) to gain deeper insights. In the conventional approach, this often begins with tasks such as:

• reading the dataset with pandas,

• performing sampling,

• generating summaries,

• and creating basic plots, etc.

While effective, these steps can be somewhat repetitive, and, if I may say, a little tedious.

The Conventional Way (Tedious and Repetitive)

show / hide

python

import pandas as pd
df = pd.read_csv(f'{base_path}/telco.csv')
df.sample(10)

	customerID	gender	SeniorCitizen	Partner	Dependents	tenure	PhoneService	MultipleLines	InternetService	OnlineSecurity	...	DeviceProtection	TechSupport	StreamingTV	StreamingMovies	Contract	PaperlessBilling	PaymentMethod	MonthlyCharges	TotalCharges	Churn
285	6202-DYYFX	Female	0	No	No	22	Yes	Yes	Fiber optic	No	...	No	No	No	No	One year	Yes	Credit card (automatic)	76.00	1783.6	No
4323	3707-LRWZD	Female	0	No	No	32	Yes	Yes	Fiber optic	No	...	No	No	Yes	No	One year	No	Electronic check	84.05	2781.85	Yes
5302	9700-ISPUP	Female	0	Yes	Yes	10	Yes	No	DSL	No	...	No	Yes	No	Yes	Month-to-month	Yes	Electronic check	65.50	616.9	No
171	1875-QIVME	Female	0	Yes	No	2	Yes	Yes	Fiber optic	No	...	Yes	Yes	Yes	Yes	Month-to-month	Yes	Electronic check	104.40	242.8	Yes
6739	6994-KERXL	Male	0	No	No	4	Yes	No	DSL	No	...	No	No	No	Yes	Month-to-month	Yes	Electronic check	55.90	238.5	No
868	3313-QKNKB	Male	0	Yes	No	59	Yes	Yes	Fiber optic	No	...	No	No	No	Yes	One year	No	Electronic check	85.55	5084.65	Yes
5102	9070-BCKQP	Female	0	Yes	No	72	Yes	Yes	Fiber optic	Yes	...	Yes	Yes	Yes	Yes	Two year	Yes	Credit card (automatic)	110.15	7881.2	No
6444	6302-JGYRJ	Male	0	No	Yes	31	Yes	Yes	DSL	No	...	No	Yes	Yes	Yes	One year	Yes	Mailed check	79.45	2587.7	Yes
3137	4567-AKPIA	Female	0	Yes	Yes	41	Yes	Yes	Fiber optic	No	...	Yes	Yes	Yes	Yes	One year	Yes	Bank transfer (automatic)	109.10	4454.25	No
6360	8073-IJDCM	Female	0	No	No	1	Yes	No	No	No internet service	...	No internet service	No internet service	No internet service	No internet service	Month-to-month	Yes	Mailed check	20.30	20.3	Yes

10 rows × 21 columns

Automate it! (Minimal Effort, Maximal Gain)

show / hide

python

from ydata_profiling import ProfileReport
profile = ProfileReport(df, title="Profiling Report")
profile.to_notebook_iframe()

Auto Eda

Data Preprocessing

show / hide

python

from sklearn.pipeline import Pipeline
from sklearn.compose import ColumnTransformer
from sklearn.impute import SimpleImputer
from sklearn.preprocessing import StandardScaler, OrdinalEncoder, OneHotEncoder, LabelEncoder

ids_cols = ['customerID']
num_cols = ['MonthlyCharges', 'tenure']
txt_cols = ['TotalCharges']
tgt_cols = ['Churn']
cat_cols = list(set(df.columns) - set(ids_cols) - set(num_cols) - set(txt_cols) - set(tgt_cols))

numeric_transformer = Pipeline(steps=[
    ('imputer', SimpleImputer(strategy='mean')),
    ('scaler', StandardScaler())
])

categorical_transformer = Pipeline(steps=[
    ('imputer', SimpleImputer(strategy='most_frequent')),
    ('encoder', OrdinalEncoder())
])

preprocessor = ColumnTransformer(transformers=[
    ('num', numeric_transformer.set_output(transform="pandas"), num_cols),
    ('cat', categorical_transformer.set_output(transform="pandas"), cat_cols)
])

X = preprocessor.fit_transform(df.drop(columns=['Churn'], axis=1))
y = LabelEncoder().fit_transform(df['Churn'])

Model Building

show / hide

python

from sklearn.model_selection import train_test_split
from sklearn.metrics import confusion_matrix, classification_report, accuracy_score
from sklearn.linear_model import LogisticRegression

X_train, X_test, y_train, y_test=train_test_split(X, y, test_size=0.25, random_state=42)
LR = LogisticRegression()

model = LR.fit(X_train, y_train)
y_pred = model.predict(X_test)

print(classification_report(y_test, y_pred))

              precision    recall  f1-score   support

           0       0.85      0.90      0.88      1282
           1       0.69      0.57      0.63       479

    accuracy                           0.81      1761
   macro avg       0.77      0.74      0.75      1761
weighted avg       0.81      0.81      0.81      1761

Low Code Machine Learning with PyCaret

As we can see in the Data Preprocessing and Model Building steps, these often involve repetitive workflows. At this point, we should ask ourselves: Is it really necessary to perform every step manually? Do these steps directly help us answer our main objective, in this case, preventing more customer churn, or could we use more automated tools as long as they allow us to achieve the same goal effectively?

show / hide

python

from pycaret.classification import *
exp = ClassificationExperiment()
exp.setup(df.drop(columns=['customerID'], axis=1), target = 'Churn', session_id = 123)
best = exp.compare_models()

	Model	Accuracy	AUC	Recall	Prec.	F1	Kappa	MCC	TT (Sec)
nb	Naive Bayes	0.7789	0.8060	0.7789	0.7727	0.7741	0.4110	0.4139	0.0830
knn	K Neighbors Classifier	0.7716	0.7792	0.7716	0.7642	0.7667	0.3912	0.3932	0.0880
qda	Quadratic Discriminant Analysis	0.7576	0.7972	0.7576	0.7248	0.7100	0.2137	0.2494	0.0830
lr	Logistic Regression	0.7550	0.8296	0.7550	0.7408	0.6890	0.1566	0.2285	0.1650
rf	Random Forest Classifier	0.7501	0.7778	0.7501	0.7130	0.6983	0.1819	0.2174	0.1780
svm	SVM - Linear Kernel	0.7462	0.7937	0.7462	0.7132	0.6917	0.2081	0.2471	0.0870
ridge	Ridge Classifier	0.7385	0.7839	0.7385	0.6907	0.6592	0.0793	0.1278	0.0830
lda	Linear Discriminant Analysis	0.7385	0.7338	0.7385	0.6907	0.6592	0.0793	0.1278	0.0780
et	Extra Trees Classifier	0.7379	0.7473	0.7379	0.6890	0.6710	0.1069	0.1467	0.1600
lightgbm	Light Gradient Boosting Machine	0.7349	0.7311	0.7349	0.7270	0.7271	0.2888	0.2942	0.2100
dummy	Dummy Classifier	0.7347	0.5000	0.7347	0.5398	0.6223	0.0000	0.0000	0.0920
ada	Ada Boost Classifier	0.6848	0.7266	0.6848	0.7221	0.6956	0.2707	0.2786	0.1230
gbc	Gradient Boosting Classifier	0.6777	0.6702	0.6777	0.6892	0.6690	0.1716	0.1857	0.2200
dt	Decision Tree Classifier	0.5262	0.5046	0.5262	0.6230	0.5306	0.0122	0.0212	0.1240

Business Intelligence Certification

For Business Intelligence certification, the required units include:

• Data Collection

• Data Exploration

• Data Validation

• Data Definition

• Data Construction

• Data Integration

• Dashbord Development

As of this case, I also believe they could be simplified, such as:

• Data Collection

• Data Understanding

• Business-Centric Objective

• Dashboard Development

• Dashboard Deployment

Several steps, such as Data Integration, are omitted here, since in my view they align more closely with the responsibilities of data engineering rather than Business Intelligence.

Business-Centric Objective

We have covered part of Data Collection and Understanding before, now comes to the most important aspect of business intelligence, that is define the objective constrained by business-centric. This part will be heavily inspired by the Google Looker Guide 'The Art of Telling Stories With Data' that ca be accessed [2].

Choosing the Right Visualization

Who is your audience?

• Is it an executive who is interested in overall business performance?

• Is it the sales team that wants to determine whether or not they will hit their quota for the quarter?

• Is it the marketing team that needs to determine if their social media and email campaigns are effective?

What is your goal?

• Are you building something that is referential, such as a report? Or a visualization that’s meant to A CEO or CFO may want to understan how revenue this year compares to last year and the year before that.

• Are you creating something that is operational, like a dashboard that people will continually check to help direct their workflow?

What are you trying to show or say?

• Are you comparing data across different categories (bar chart)?

• Are you highlighting changes over time (line chart)?

• Are you presenting a picture of the full data set (scatterplot or histogram)? A clear understanding of all these factors will help you zero in on a visualization that effectively speaks to your audience.

Different Users Have Different Questions and Require Different Data

Data Granularity

Give Context

Context First

Choosing The Right Chart

Appropriate Chart

Use Case

Example Case

PS

The example use case (BI dashboard) mentioned earlier [3]. was built using Tableau. Within the Business Intelligence ecosystem, there are many tools available, ranging from closed-source to open-source solutions.

What I would like to emphasize is that tools are just tools. As practitioners, we have the freedom to choose any technology stack, as long as it enables us to deliver insights and effectively address the business problem. You can even rely on something as simple as Excel, or take advantage of emerging Generative AI/LLM-based tools, such as LIDA [4].

Ultimately, while tools will continue to evolve, the fundamentals of Business Intelligence remain the same.