Developed a custom evaluation metric called the Age Prediction Metric to predict the ages of placebo vs subjects synthetically (Using GANS) subjected to facial treatments.

• 25% increase in accuracy

  Shifting from Tree Based Models to a Custom ANN with Batch Processing, Dropouts and Custom Callbacks such as Learning Rate Decay and Early Stopping helped the model converge faster and increase the overall F1 Score (Accuracy) by 25%.

• 30% decrease in training time

  Utilized a combination of feature selection techniques, namely SHAP, Pearson's Correlation, and Chi-Square Tests, on a vast dataset to effectively eliminate highly correlated features and unique identifiers. This strategic approach substantially reduced the dimensionality of the data, leading to significant time savings during the training process.