TeamZero

TeamZero

Inspiration

Our journey began with a vision to revolutionize the insurance sector, inspired by the stories we heard from countless individuals facing complexities in their insurance processes. We were moved by the idea of simplifying these experiences, making them more transparent and personalized, and ultimately providing peace of mind to policyholders.

What it does

Our project harnesses the power of XGBoost and Optuna within the realm of data analytics to predict customer satisfaction and conversion rates. By doing so, we aim to enhance the customer journey, streamline the application process, and tailor communication to individual needs, ensuring that potential policyholders stay engaged and informed.

How we built it

Over the past 2-3 days, we have conducted extensive data analysis, data pre-processing, model training steps to predict the outcomes of f_purchase_lh using Python. We built our project on the pillars of machine learning, utilizing Python and its robust libraries. There were a total of 304 columns in the parquet file provided by Singlife, which contained 3 different dtypes: float64(44), int64(46) and object(214). There are 3 main evaluation metrics: Precision, Recall and F1-score and we have set the optimization of the F1-Score as our quantitative priority. Next, we leveraged Optuna for hyperparameter tuning of our XGBoost model, focusing on maximizing the F1 score. The process involved extensive experimentation with different parameters and validation techniques to ensure our model's robustness.

Challenges we ran into

The road was not without its bumps. We faced challenges with data imbalance, which skewed our initial predictions. Tuning hyperparameters for the best model performance required numerous iterations and a delicate balance between precision and recall. Another hurdle was ensuring data privacy and security while handling sensitive customer information.

Accomplishments that we're proud of

We are proud of developing a model that significantly enhances prediction accuracy, evidenced by improved AUC and precision scores. We managed to implement a more refined data cleaning process and sophisticated feature engineering, which were crucial in improving our model's performance.

What we learned

This project has been a profound learning curve for us. We've gained a deeper understanding of the nuances of machine learning models and the importance of hyperparameter tuning. We've also learned that data quality is as crucial as the algorithms we employ. Most importantly, we've learned to persevere, collaborate effectively, and think critically to overcome challenges.

What's next for TeamZero

TeamZero is just warming up. Looking ahead, we aim to integrate real-time analytics, explore the potential of neural networks, and consider deploying our model into a live environment. We're excited to continue refining our system, always to put the client's needs at the forefront of technological innovation. Therefore other than Data Analysis, we require strong Design Thinking methodologies to ensure a curated and suitable solution for clients.

Team: Reina, Ryan, Bowen, Claudia