Analyzing Direct Marketing Campaigns for Term Deposits in a Portuguese Banking Dataset
Financial institutions depend heavily on direct marketing strategies, and evaluating their effectiveness can provide insightful information. The dataset at the disposal goes into a Portuguese banking institution’s direct marketing methods, specifically phone call campaigns focused on boosting term deposits. These campaigns frequently required many contacts with clients to determine their term deposit product subscription.
Dataset Description
Key Features:
· Age: Represents the age of the clients involved.
· Job: Categorizes clients based on their occupations.
· Marital Status: Indicates the marital status of the clients.
· Education Level: Reflects the educational background of the clients.
· Default: Indicates if the client has credit in default.
· Balance: Represents the average yearly balance of the clients in euros.
· Housing and Loan: Indicates if the client has housing or personal loans.
· Contact: Specifies the communication type used (cellular or telephone).
· Day and Month: Capture the timing of the last contact made.
· Duration: Reflects the duration of the last contact, influencing the subscription decision.
· Campaign: Represents the number of contacts made during the campaign.
· Pdays: Signifies the number of days since the client was last contacted.
· Previous and Poutcome: Detail the previous marketing campaign’s outcomes.
Target Variable:
· y (Term Deposit Subscription): Binary variable denoting whether the client subscribed to the term deposit or not.
As this dataset does not contain any missing values and contains plenty of information, we can directly jump to the analysis part in PredictEasy without doing data cleaning. Important note: The duration attribute highly affects the output target (e.g., if duration=0, then y=‘no’), so we will remove this variable from the dataset.
PredictEasy Analysis
In a previous blog article (Using Classification Analysis to Improve Efficiency and Sustainability in Conventional Power Plants), I used PredictEasy to improve the efficiency of power plants. This blog will use the same tool to dive deep into this banking dataset to get insights to improve term deposit subscriptions.
We start by putting every variable in X and the target variable in Y. After doing this, we see the summary:
As we can see, there is a correlation between previous and pdays, so we can remove one of the variables to fine-tune in the next iterations (we will be eliminating pdays)
From the feature rank graph, age, marital and education factors do not contribute much, so we remove them in the next iterations.
Fine-Tunning
After removing unimportant features, we again ran PredictEasy, and this time, we got to see this:
After seeing the SHAP plot and Feature Rank graph, we learned that month, campaign, default, previous, balance, and poutcome do not contribute.
Again, by removing the unimportant variable, we get our final model with an accuracy of 0.87 and a precision of 0.81.
What-If Analysis in PredictEasy
Through what-if analysis, we get the idea of the result of the model if we put the value of our choice. Through this, we can verify whether the model works correctly or incorrectly.
If the day values are at the middle or last of the month, then the term deposits subscriptions is yes, but if it is in the first ten days, it is no.
We also learned that if we change the value of the loan and housing to yes, then subscriptions turn to no.
Conclusion
In conclusion, the analysis we performed revealed some of the major trends driving term deposit subscriptions:
The best time to subscribe is the middle of the month.
Financial Status: Higher subscription rates are seen by clients who have no loans and no housing value.
Job Impacts: The likelihood of subscribing is higher for administrative, entrepreneur, and retired jobs.
These data provide insightful advice on focusing marketing efforts and modifying tactics to maximize term deposit subscriptions. Knowing these trends helps banking organizations create more successful campaigns that produce better results.
Dataset Source
Moro,S., Rita,P., and Cortez,P.. (2012). Bank Marketing. UCI Machine Learning Repository. https://doi.org/10.24432/C5K306.
Further Reading
Decoding the Endgame: Navigating Tic-Tac-Toe’s Final Moves by Elsa
A Predictive Analysis of CHP in Power Plants by Marsel