Home

Final Model

What you'll see here

Why M8 was selected, how it performs on the untouched 5,000-row test set, and what that is worth in avoided maintenance cost.

Test accuracy
98.74%
5,000 unseen units
Test recall
87.59%
247 of 282 failures caught
Test precision
89.82%
28 false alarms
Test F1 · ROC-AUC
88.69% · 0.939
balanced and well ranked
Final model selection — why M8
CandidateVal recallVal F1Val precisionVerdict
M5 Adam + dropout90.99%93.95%97.12%Best F1, but misses more failures
M6 SGD + class weights91.44%75.75%64.65%Recall bought with 1-in-3 false alarms
M10 Adam + dropout + OS91.44%88.26%85.29%Ties on recall, overfits the duplicated rows
M8 Adam + dropout + CW91.89%91.48%91.07%Selected — highest recall, precision held
  • Recall is the primary metric, and M8 leads it while keeping precision above 91% — every extra failure caught does not come at the cost of a flood of inspections.
  • Its training and validation recall are identical at 91.89%, so the model generalises rather than memorises.
  • Class weights are preferred over oversampling because they reweight the loss without duplicating rows.
Architecture

Adam (lr 0.001) · Binary cross-entropy with class weights · 30 epochs · batch 128 · seed 42 · threshold 0.5

Input
40 standardised predictors
Dense 128
ReLU, He-normal init
Dropout
0.3
Dense 64
ReLU
Dropout
0.3
Dense 32
ReLU
Dropout
0.3
Output
1 unit, sigmoid
Confusion matrix (test set)
Predicted healthyPredicted failure
Actual healthy4,69028
Actual failure35247
  • 247 failures caught, 35 missed
  • 28 false alarms, 4,690 healthy units cleared
  • ROC-AUC 0.939
  • Validation recall was 91.89% and test recall is 87.59% — a small, expected drop on genuinely unseen data with no sign of overfitting.
Business impact — illustrative cost scenario

Unit costs are assumed ($5k inspection / $15k repair / $60k replacement); the project statement supplies only the ordering replacement > repair > inspection

No model
$16.92M
every one of 282 failures becomes a replacement
With the model
$5.93M
247 repairs, 35 replacements, 28 inspections
Cost avoided
$10.98M
64.9% below baseline

Unit costs are illustrative and preserve the stated ordering; the percentage saving is what should be read, not the absolute dollars.

Threshold tuning — recall, precision and expected cost

Real M8 test-set probabilities on 5,000 unseen generators. Move the threshold to see the operating point change live.

threshold 0.50
0.05 — catch nearly everything, flood inspections0.95 — near-certain alarms only, more misses

Recall

87.59%

247 caught · 35 missed

baseline

Precision

89.82%

28 false alarms

baseline

F1 score

88.69%

accuracy 98.74%

baseline

Expected cost

$5.95M

64.9% below the no-model baseline

baseline

At 0.50 the model catches 247 of 282 failures, raises 28 false alarms and costs $5.95M versus $16.92M with no model.

Expected cost = 247 repairs × $15,000 + 35 replacements × $60,000 + 28 inspections × $5,000.

Cost is flat between roughly 0.35 and 0.70 — the operating point is robust, so 0.50 is kept for interpretability while a lower threshold stays available whenever missed failures become more expensive.

AI: Final model interpretation

Click Generate to produce an evidence-based observation, insight, recommendation, and business benefit for this section.