Roshai

I appeared for the RoshAI interview process, which primarily focused on machine learning fundamentals, model evaluation, and practical problem-solving. The interviewer emphasized understanding the reasoning behind model selection and the ability to apply machine learning concepts to real-world scenarios. The discussion began with questions on choosing between Logistic Regression, Random Forest, and Neural Networks for different problems. I explained the factors influencing model selection, including dataset size, feature complexity, interpretability, computational cost, and the nature of the relationships within the data. This was followed by questions on why accuracy can be a misleading metric for imbalanced datasets, where I discussed alternative metrics such as precision, recall, F1-score, and ROC-AUC. The interviewer then explored topics related to model performance and generalization. I was asked how to detect and reduce overfitting, where I explained techniques such as cross-validation, regularization, pruning, early stopping, and increasing training data. Another discussion focused on the bias-variance tradeoff and how balancing underfitting and overfitting influences model selection. I was also asked when precision is more important than recall, and vice versa, using practical examples such as spam detection and disease diagnosis. The interview also covered data preprocessing and feature engineering. I explained how to handle missing values when they are not missing at random, the concept of data leakage and its impact on model validity, and why feature scaling is essential for algorithms such as Logistic Regression and Support Vector Machines but generally unnecessary for tree-based models. The interviewer also asked me to describe the evaluation metrics used in my projects, where I explained the rationale behind selecting different metrics based on the problem statement. Finally, the discussion moved toward information retrieval, where I was asked about rank fusion algorithms and how combining rankings from multiple retrieval methods can improve overall search quality. Overall, the interview was concept-driven and focused on evaluating practical machine learning knowledge, model evaluation strategies, and the ability to justify technical decisions with clear reasoning.