Projects
Data Science & Machine Learning
Projects
E-Commerce Customer Churn Analysis to Reduce Acquisition and Retention Costs using Predictive Modeling
July 2024
The project addressed a common business issue where an e-commerce focus on acquiring new customers over retaining existing ones. The task was to develop a solution to identify customers at risk of churn by analyzing customer behavior data. A predictive model was built with a recall of 94.04% and precision of 89.77%, minimizing false negatives and accurately identifying at-risk customers. This allowed the company to implement targeted retention strategies, significantly improving customer retention and reducing costs by 79.45%.
Bike Demand Prediction to Optimize Resource Allocation and Reduce Operational Cost
May 2024
The project aimed to address the challenge of predicting hourly bike rental demand in a bike-sharing system, influenced by factors like weather, season, and time. The task involved building a predictive model to optimize resource allocation and reduce operational costs. The model, developed using XGBRegression, achieved strong performance metrics, including an MAE of 23.125 and MAPE of 29.06. By implementing this model, the company could reduce operational costs by 10% and increase daily customers by 15%. Additionally, with dynamic pricing, the daily profit increased by 24.375%.
Comprehensive Analysis on Airbnb Listings in Bangkok to Increase Host & Airbnb's Revenue
April 2024
The project involved analyzing factors that affect the pricing and popularity of Airbnb listings in Bangkok. The challenge was to understand how various characteristics, such as room type, location, and language used in the listing title, influence daily rates and listing popularity. The task was to identify these key factors and provide actionable recommendations to optimize pricing and boost the profitability of listings. Through detailed analysis, it was found that central locations, short-term rentals, and English-language titles command higher prices and popularity. Recommendations were made to enhance facilities and services, implement dynamic pricing, and offer discounts for long-term stays. These strategies have the potential to increase revenue for both hosts and Airbnb by up to 28.771%.
Enhancing Loan Risk Management through Predictive Analysis
September 2023
The project involved working with a large loan application dataset containing over 466,000 entries. The goal was to build a predictive model that could accurately identify potential loan defaults, prioritizing a high recall score to minimize financial losses. Several machine learning models, including Logistic Regression and GradientBoost, were applied after thorough data cleaning, feature engineering, and balancing the dataset. The Logistic Regression model was chosen for its performance, achieving a recall score of 0.78 and an AUC of 0.848, effectively reducing the risk of misclassifying bad loans.
Data-Driven Analysis of Loan Default Risk for Home Credit
August 2023
This project aimed to build a model to predict the risk of loan default for Home Credit, focusing on applicants with limited credit histories. Using data from multiple sources, the process involved feature selection and model building with Logistic Regression and LightGBM algorithms. The model achieved an AUC score of 0.785, improving the accuracy of credit risk assessment. These insights enable Home Credit to make better-informed decisions regarding loan approvals, ultimately helping reduce the default rate and manage financial risk more effectively.
Deep Learning
Projects
Detecting Cyberbullying with Advanced NLP Models: A Comparative Study of ERNIE, XLNet, and RoBERTa
June 2024
The project involved building and evaluating Natural Language Processing (NLP) models to classify various types of cyberbullying, such as gender, religion, and ethnicity-related bullying, using a labeled dataset. After thorough data preprocessing, models like ERNIE, XLNet, and RoBERTa were trained and validated. RoBERTa showed the best performance with 85.09% accuracy but struggled with overfitting. The analysis identified key improvements like applying regularization techniques, using smaller batch sizes, or even use a simpler machine learning models to enhance model generalization and reduce overfitting, while the overall goal was to accurately classify cyberbullying instances
Exploring Deep Learning Models for Polyp Segmentation: A Comparative Analysis
January 2024
In this project, deep learning models were explored for colorectal polyp segmentation using the Kvasir-SEG dataset. The objective was to compare the performance of various models like FCN8, FCN16, FCN32, PEFNet, and DeepLabV3+ for accurate polyp segmentation. After preprocessing and model training, the evaluation showed that PEFNet achieved the highest accuracy of 96.07% with a mean Dice coefficient of 89.71%. However, the trade-off was its longer computational time, highlighting the balance needed between accuracy and efficiency for clinical applications.
Statistical
Projects
Forecasting the Monthly Consumer Price Index of DKI Jakarta: Integrating Predictive Analytics for Enhanced Economic Growth Strategies
June 2024
This project focused on improving the accuracy of predicting Jakarta's Consumer Price Index (CPI) by analyzing CPI data from 2009 to 2023. Using R programming language, several forecasting models, including Naivë, Holt-Winters, SARIMA, and Neural Networks, were evaluated. The Neural Networks model emerged as the most effective, achieving the lowest Mean Absolute Error (MAE) of 0.798, Root Mean Squared Error (RMSE) of 1.041, and Mean Absolute Percentage Error (MAPE) of 0.721 on the testing dataset. These findings demonstrate that this model successfully captured seasonal patterns and trends, providing reliable forecasts to support informed economic policy decisions.
Assessing the Relationship between Educational Attainment and Social Well-being in Indonesia through Canonical Correlation Analysis
June 2024
This project aimed to explore the correlation between educational attainment and socio-economic well-being in Indonesia using Canonical Correlation Analysis (CCA) with R programming language. Data from the Indonesian Central Bureau of Statistics for 2023 were analyzed, revealing that high school completion rates and literacy rates had a significant impact on socio-economic indicators. The first canonical function demonstrated a strong correlation of 0.924, with 85.5% of the variance in socio-economic outcomes explained by educational metrics. This result highlighted the pivotal role of education in reducing poverty and improving unemployment and Human Development Index scores. These findings provide key insights for policymakers to address disparities and promote inclusive growth in Indonesia.
Promoting Sustainable Economic Growth: A WLS Approach to Address Heteroscedasticity in Predicting Technological Literacy Rate
January 2024
With the use of R programming language, this project focused on predicting the technology literacy rate in 34 provinces of Indonesia by addressing the issue of heteroscedasticity using the Weighted Least Squares (WLS) method. We analyzed key factors such as poverty, the non-food poverty line, and education levels, finding that poverty had a negative impact on technological literacy, while factors like the non-food poverty line and human development index (HDI) had positive correlations. The WLS model showed an improved adjusted R-squared, AICc, and BIC (0.8826, 206.0274, 211.7258, respectively) indicating better explanatory power compared to the initial model which was 0.8786, 268.8353, 214.0989, respectively. These insights highlight crucial socio-economic factors for policymakers to address in efforts to boost technology literacy and support sustainable economic growth.
Programming
Project
EcoHaul: A Web Platform Bridging Consumers and Recycling Services for a Cleaner Environment
June 2024
This project involved developing EcoHaul, a Laravel-based web application designed to bridge the gap between consumers and recycling services. Recognizing the growing need for effective waste management, the goal was to provide users with seamless access to features such as pick-up and drop-off services, as well as informative newsletters to raise environmental awareness. By implementing these functionalities, the project aimed to encourage greater participation in recycling efforts. The application successfully introduced features like user registration, service requests, and eco-reward points, resulting in increased user engagement and contribution toward a cleaner environment.