Team 265

Inspiration

There are many classification problems in our daily life. This datathon encouraged us to derive actionable insights from this data to enhance the customer experience. We are inspired to search for more data cleaning and modelling techniques.

What it does

Our project focuses on analyzing Singlife's dataset to uncover insights that can enhance the customer journey. We aim to identify critical touchpoints that contribute to customer drop-off and find opportunities to streamline the application process.

How we built it

We built our project by first thoroughly examining Singlife's dataset. This involved cleaning and preprocessing the data to ensure that data within a certain range of variance are selected. We then employed advanced analytics and machine learning techniques to extract meaningful patterns and insights. The process included identifying key touchpoints, analyzing customer behavior, and creating predictive models for customer satisfaction and conversion rates. We utilized a combination of data science tools and programming languages to implement our solutions.

Challenges we ran into

Choice of Model: high accuracy with relatively low recall Data cleaning: some categorical data has to be converted into numerical value, which then can be taken as an input to our model Imbalanced data sample: we observe that for some of the features, majority of them appear to be either 0 or NA, thus we adopted some data manipulation methods such as SMOT to have a balanced data input to our model

Accomplishments that we're proud of

For data cleaning, we are able to filter out irrelevant non-numerical data. In the end, we are able to achieve an accuracy around 90 percent, with a recall around 0.25, using random forest with a grid search method. We also adopted few other data modelling such as Knn neighbour classification, logistic regression and balanced bagging.Eventually, we are able to come up with top 5 models optimised to the similar performance, and selected random forest with grid search as the most stable one.

What we learned

As a computer science student who never tried data processing before, I learned the basic workflow of data analysis and modelling and tried different basic data pre-processing techniques. As a team, we learnt how to collaborate using git and we all learnt new techniques for modelling.