In our advanced machine learning course, I embarked on a project focused on estimating a person's age based on neurodata. This task holds great significance as it enables the detection of accelerated aging, a crucial aspect in healthcare. Our project involved working with a table containing numerous anatomical features of a thousand different brains. However, not all features were meaningful, and including irrelevant data in machine learning can lead to inaccurate outcomes.
A substantial portion of our project involved data preparation. I employed statistical methods, such as ANOVA or the Pearson coefficient, to identify meaningful variables that correlated with age. I also addressed the challenge of missing data, as it didn't imply uselessness.
For age estimation, I utilized a boosting regression method, which balanced variance and bias optimally. This method also prevented overfitting to the dataset, ensuring its performance on unseen data. Fine-tuning the boosting method was necessary, typically achieved by defining a search area. However, this computational-intensive process required several hours of computation on a computer.
Through our project, I gained insights into the intricate task of estimating age from neurodata, highlighting the importance of data preparation, feature selection, and appropriate modeling techniques.