Embedded Files
Loan-default Prediction
Banks earn a major revenue from lending loans. But it is often associated with risk. The borrowers may default on the loan. To mitigate this issue, the banks have decided to use Machine Learning to overcome this issue. They have collected past data on the loan borrowers & would like you to develop a strong ML Model to classify if any new borrower is likely to default or not.
This project aims to develop a machine learning model that helps banks predict loan default risk for new applicants. By utilizing historical borrower data, the model will classify applicants as likely to default or not, enabling banks to make informed lending decisions and minimize potential losses.
Modeling Claim Severity for Motor Liability Insurance
Car liability insurance provides financial protection for your vehicle against collisions. To get insured you purchase a policy and pay a premium. In case of an incident you fill up a claim and the insurer decides to match the claim amount. But what is the fair amount of premium? Usually, the amount of premium is determined by considering factors such as driving age, vehicle brand, area in which one lives in, etc. In this project, the fair amount of premium is investigated by considering a French motor third-party liability (MTPL) insurance portfolio with corresponding claim counts and amounts observed in the period of exposure. The project involves data exploratory data analysis, feature engineering, modeling data by decision trees, and investigating model performance. Further conclusions and improvements are discussed in the end.
Short-Term Energy Demand Forecasting
This case study addresses the challenge of short-term energy demand forecasting using hourly electricity consumption data. The objective is to generate accurate one-day-ahead forecasts to support operational decisions in energy supply and trading.
Several time series models were evaluated, including:
ETS (Exponential Smoothing) as a baseline,
SARIMA for handling trend and seasonality,
RNN with LSTM layers for capturing nonlinear temporal patterns.
Model selection and preprocessing were guided by exploratory data analysis. An additive ETS model with damped trend and daily seasonality was chosen based on observed patterns. Differing techniques were applied to achieve stationarity, improving model interpretability and performance.
Forecast accuracy was evaluated using RMSE, MAE, and MAPE. A rolling-origin evaluation with stepped windows was used for efficient, realistic validation. The study highlights the effectiveness of simple, well-structured models and points to future improvements through external data integration and deeper model tuning.
Page updated
Google Sites
Report abuse