Livoa LogoLivoa
Parameter
Formula / Logic
Notes
Acceleration magnitude
√(ax2 + ay2 + az2) / 9.81
Used to detect spikes for falls
Fall detection
Spike > 2.5g + inactivity <0.5g for 2s
Reduces false alarms
Pulse (BPM)
BPM = beats in 10s × 6 (using peak detection)
More accurate than linear mapping
Body temp
>38°C or <35°C
Threshold alert
Ambient temp
<15°C or >35°C
Threshold alert
Humidity
<20% or >80%
Threshold alert
✔
Summary of Thresholds
Parameter
Current Threshold
Notes
Pulse
<50 or >120 bpm
Adjust based on age/activity
Body Temp (°C)
<35 or >38
Fever/hypothermia alert
Ambient Temp (°C)
<15 or >35
Cold/heat warning
Humidity (%)
<20 or >80
Very dry/high humidity
Fall (accel g)
>2.0
Adjust for sensitivity
Model
Detection Acc. (%)
Classification Acc. (%)
Linear Regression
92
48
Logistic Regression
95
52
Polynomial Regression (4th deg.)
94
60
MLP Classifier
97
72
Naïve Bayes
88
45
Decision Tree
96
70
SVM
95
58
KNN
93
55
1D-CNN (Proposed)
99
98
Model
Detection Accuracy
Classification Accuracy
Remarks
Linear Regression
0.02
0.10
Performs poorly due to inability to capture non-linear fault dynamics.
Logistic Regression
0.82
0.25
Effective for binary fault detection but weak in multi-class fault separation.
Polynomial Regression
0.85
0.84
Handles moderate non-linearities but prone to overfitting.
MLP Classifier
0.18
0.12
Shallow neural network underfits due to limited depth.
Naive Bayes
0.82
0.40
Works well for independent features; limited for correlated current–voltage data.
Decision Tree
0.95
0.89
Best among classical models; strong interpretability and quick learning.
SVM
0.78
0.23
Requires kernel tuning; struggles with overlapping class boundaries.
KNN
0.65
0.41
Performs moderately; accuracy drops with high feature dimensions.
1D-CNN (Proposed)

0.98

0.98

Achieves highest accuracy due to deep temporal feature extraction.

PROJECT PHASE

by AAKASH

0
0 uses