Indian Cuisine Classification Framework — A Culturally Contextualized Image Dataset for Statistical Learning Education

We present the Indian Food Image Dataset and a curriculum specifically developed for instructing statistical learning in undergraduate and graduate statistics and data science courses. The dataset consists of 4,577 RGB images categorized into 100 regional Indian food groups and was created with Pedagogy as a key design criterion: its medium size, culturally relevant content, and detailed class structure make it a good fit for active learning in the classroom. Using this dataset, students face real statistical problems like generalization error, the bias-variance trade-off, class imbalance, and multiclass evaluation metrics (precision, recall, F1-score, and confusion matrix analysis) in a field that is interesting and relevant to a lot of students around the world.This paper presents not only the dataset and a hybrid CNN-Transformer benchmark (top-1 test accuracy 89.08%) but, more significantly, four meticulously detailed lesson plans that include learning objectives, sequential instructor guidance, estimated timing, and exemplar student deliverables. We detail a pilot implementation in an advanced data science course at a university in India, presenting observations of student engagement and assessment results. Anyone can access all of the data, the Python code that can be reused, and the Jupyter notebook exercises. The dataset and curriculum meet a need in statistics education by making available a real-world, open, culturally relevant classification problem directly tied with teaching materials that are ready to be used.