AI Application Domains
Domain 8 - AI Application Domains. This domain tests whether you can map a real problem to an appropriate AI technique and justify the match, sector by sector, using data type and label availability as the deciding evidence.
Core rules and concepts
- Healthcare: medical imaging classification, triage support, drug discovery, epidemic forecasting. Constraints: patient privacy, regulatory approval, clinician oversight.
- Agriculture: crop-disease detection from leaf images (CNN classification), yield prediction (regression), irrigation scheduling from sensor data, pest monitoring by drone.
- Education: adaptive learning paths, automated short-answer scoring, dropout-risk prediction, plagiarism detection.
- Finance: fraud detection (imbalanced classification), credit scoring, algorithmic trading, anti-money-laundering pattern mining.
- Transport and manufacturing: ADAS perception and lane detection, route optimisation, predictive maintenance from sensor time series, defect detection on production lines.
- Public services and generative AI: chatbots, translation, summarisation and content generation, with hallucination, copyright and misinformation risks.
- Matching rule: image → CNN; sequence or time series → RNN/LSTM or lag-feature boosting; tabular → decision trees and ensembles; unlabelled grouping → clustering; fully deterministic rule → no ML needed.
Worked example
A block wants to detect paddy leaf disease from phone photographs. Input is an image, output is one of three classes (healthy, blast, blight), and labelled examples exist, so the match is supervised image classification using a CNN or transfer learning. Justify by data type (image), label type (discrete class) and data volume. Contrast this with switching on a pump when soil moisture falls below a fixed threshold: that is a deterministic rule and needs no learning at all.
Traps that cost marks
- Proposing deep learning for a five-row dataset; a rule or simple model is honest.
- Confusing classification (disease present or not) with regression (expected yield in kg).
- Ignoring deployment reality: connectivity, device cost, local language, maintenance.
- Overclaiming that AI replaces doctors, teachers or agronomists rather than supporting them.