Our team of three embarked on a two-week project to develop a predictive model for new loan applications using real-world data consisting of thousands of observations and 44 features. We started from scratch and successfully created a model integrated with a user interface.

The primary objective was to assess the eligibility of individuals for loans and determine the maximum loan amount they could receive based on the personal information provided through a website form. To accomplish this, we employed Streamlit to construct the website and execute the model, utilizing the input features provided by the users. Leveraging Streamlit, we constructed a user-friendly website that integrated our model and utilized user-provided input features.

Throughout the project, we faced some challenges that we successfully overcame with collaborative efforts, meticulous feature engineering, and model fine-tuning.

Problems faced

During our project, we faced two main challenges. Firstly, the masking of real people's information in the dataset negatively impacted the model's performance, resulting in a significant decrease in the correlation between features and the target variable. However, through careful analysis, feature engineering, and model optimization, we were able to improve the models' performance and mitigate the impact of the masked data.

Secondly, we encountered a highly unbalanced dataset, with the loan rejection being predominant while the accepted were underrepresented. This presented a challenge in training accurate models, as they tended to favor the majority class and struggled to make accurate predictions for the minority classes. To address this, we employed specialized techniques to handle class imbalance, ensuring fair and reliable predictions across all classes.

Key Responsibilities

Performed data preprocessing tasks to optimize data ingestion into Python.

Implemented data cleaning techniques.

Addressed missing values effectively based on each feature.

Utilized One-Hot Encoding to transform categorical data into a format suitable for analysis.

Applied scaling techniques to normalize numerical features based on their distribution.

Developed a robust pre-processing pipeline to streamline data transformation and preparation.

Contributed to the development of the website, ensuring seamless integration of data engineering components.