PredictIQ
Five prediction models — football, students, jobs, property and business — with the reasons behind each answer.
Overview
PredictIQ trains a decision tree for each of five domains on 10,000-row Excel datasets and lets users enter their own case. Each prediction comes with a probability and the key factors that pushed it up or down, computed from feature importance and correlation with the outcome.
Objectives
- Demonstrate explainable predictions on familiar topics
- Keep the data inspectable: every dataset is a plain Excel file
Features
- Five categories with their own input forms
- Probability of success or price band per case
- Key-factor explanations per prediction
- Model statistics: training size, train and test accuracy
Screens
Highlights
Design
- Dark, high-contrast cards with one colour per category
- Animated loading steps that describe what the model is doing
Engineering
- DecisionTreeClassifier per category (depth 6), 72–78% test accuracy on held-out data
- Trees exported to JSON and run in the browser; predictions matched scikit-learn on 1,000 of 1,000 test rows
Studio audit
Clear, well-explained predictions, but the page was unusable without its server.
Found
- Every prediction required the Flask server
- Uses the Tailwind CDN build, which is meant for prototyping
Fixed
- Exported the trained trees and added an in-browser predictor, verified identical on 1,000 rows
- API calls use relative paths so the app can be hosted in a sub-folder
Still to do
- Compile Tailwind at build time instead of loading the CDN script
Scored out of 10 on functionality, interface, visual consistency, performance, responsiveness, navigation, architecture, accessibility, error handling and professional readiness.
Demonstration
The five trained models run in your browser. With the Flask server running locally, the same page uses the server instead.
Want something like this?
Tell us what you are building. We will reply with honest questions, a clear approach and the next step.