Senior Data Scientist
Current- Employed scikit-learn's Random Forest to assess feature importance using both Gini and Permutation algorithms. Further enhanced feature selection through SHAP, resulting in a 14% improvement in model performance.- Utilized caretEnsemble to build an ML model, effectively identifying prediction and feature drift. The findings were instrumental in uncovering the root cause behind the decline in model performance.- Led a team in the creation of an interactive technical dashboard for… Show more - Employed scikit-learn's Random Forest to assess feature importance using both Gini and Permutation algorithms. Further enhanced feature selection through SHAP, resulting in a 14% improvement in model performance.- Utilized caretEnsemble to build an ML model, effectively identifying prediction and feature drift. The findings were instrumental in uncovering the root cause behind the decline in model performance.- Led a team in the creation of an interactive technical dashboard for understanding and debugging ML model issues. Utilized Rmarkdown, crosstalk, dplyr, Rshiny, and plotly to enhance transparency and problem-solving capabilities.- Applied word2vec for NLP-based feature engineering, creating word embeddings for a text variable. Employed K-means clustering to group similar features, resulting in a remarkable 33% improvement in model performance.- Utilized historical transcripts data of our customers to assess their churn probabilities. Transformed the textual data into embedding spaces and subsequently clustered these embeddings. Identified that a significant number of customers were grouped within the cluster associated with service cancellations. By calculating the Euclidean distance of each customer from this cluster's centroid, derived a churn score; specifically, a shorter distance indicated a higher likelihood of churn. Show less
