Picking a machine learning capstone topic is crucial — it needs to be scoped enough to finish in a semester, but mathematically solid enough to defend in a viva. These ideas span classic supervised learning, computer vision, and NLP.
A classification model that flags at-risk students from attendance, grades, and engagement data — a favorite with academic panels because the dataset is easy to justify.
A CNN trained on MNIST or a custom character set, wrapped in a simple upload-and-predict web app.
An imbalanced-classification project using anomaly detection or SMOTE-balanced models on transaction data.
A CNN-based leaf disease classifier for agriculture, popular for its real-world impact story.
NLP-based scoring of resumes against a job description using TF-IDF or embeddings.
A text classification pipeline (Naive Bayes or fine-tuned transformer) with a browser extension or web demo front-end.
A churn model plus an interactive dashboard for a telecom/SaaS-style dataset — strong for business-analytics electives.
A regression project with feature engineering as the real showpiece, not just the model.
Classic NLP capstone — extendable with aspect-based sentiment for extra depth.
Collaborative filtering or content-based recommender with a working front-end to demo live suggestions.
A model predicting diabetes, heart disease, or similar from structured clinical data, with clear caveats about not being a real diagnostic tool.
LSTM or Prophet-based forecasting for sales/inventory data — good for showing you can handle sequential data, not just tabular.