Folio — EditionSakina Saifuddin© 2026
2021

A journey through code, data, and intelligent systems.

Folio — EditionSakina Saifuddin© 2026
SS
01Intro02Work03What I Build04Achievements05Experience06Expertise07Contact
Case studyClick a page, swipe, or use ← → to turn

SPINE DISEASECLASSIFICATION WITHNEURAL NETWORKS

VERTEBRAL CLASSIFICATION

  • Classifies spine samples as normal or abnormal.
  • Uses six biomechanical measurements.
  • Tuned and tested on unseen samples.
85.48%test accuracy
310samples
Visit the portfolio
website at:sakina-saifuddin / works
21390
Vertebral ClassificationSakina Saifuddin

SPINE DISEASECLASSIFICATION WITHNEURAL NETWORKS

Normal or abnormal spine?
Fig. 1 — Train and test confusion matrices.1
  1. Task
  2. My work
  3. Results
  4. Outcome

Normal or abnormal spine?

  • 310 samples, six measurements each.
  • Can a neural network classify them reliably?
6measurements per sample
2
A carefully tuned network
Fig. 2 — Learning-rate experiment.3
  1. Task
  2. My work
  3. Results
  4. Outcome

A carefully tuned network

  • Stratified 80/20 split with normalization.
  • Tested learning rates, momentum and layer sizes.
  • Measured accuracy, F1 and errors.

Tools: Python · TensorFlow · Keras

4
Most unseen samples classified correctly
Fig. 3 — F1 scores for each class.5
  1. Task
  2. My work
  3. Results
  4. Outcome

Most unseen samples classified correctly

85.48%test accuracy
95.56%training accuracy
310samples
6measurements
6
Six measurements go a long way
Fig. 4 — Where mistakes remain.7
  1. Task
  2. My work
  3. Results
  4. Outcome

Six measurements go a long way

  • The network separates most samples well.
  • The train/test gap shows a small-data limit.
  • Not a medical diagnostic tool.
8
9

Technologies

Python
All experiments.
TensorFlow & Keras
The neural network.
Scikit-learn
Split, scaling and metrics.
What I did
  • Prepared the data and the split.
  • Ran the tuning experiments.
  • Evaluated on held-out data.
10

VERTEBRAL CLASSIFICATION

  • Classifies spine samples as normal or abnormal.
  • Uses six biomechanical measurements.
  • Tuned and tested on unseen samples.
85.48%test accuracy
310samples
Visit the portfolio
website at:sakina-saifuddin / works
21390
Cover