muzammil-khan/wearable-activity-classifier

Wearable Activity Classifier

A CNN-based deep learning model for classifying wearable sensor signals into three activity classes:

• Stationary
• Walking
• Running

Model Details

The model was trained as part of a deep learning laboratory activity using wearable sensor data.

Architecture

The selected model is a 1D Convolutional Neural Network:

• Conv1D: 32 filters, kernel size 3, ReLU activation
• MaxPooling1D: pool size 2
• Flatten
• Dense: 32 units, ReLU activation
• Dense: 3 units, Softmax activation

Input

The model expects:

• 100 sensor readings
• 1 sensor feature
• Input shape: (100, 1)

Output Classes

The model predicts three classes:

1. Stationary
2. Walking
3. Running

Evaluation Results

The CNN achieved the following results on the test set:

Metric	Result
Test Accuracy	100%
Parameters	50,435
Training Time	~2.9 seconds

The CNN was selected because it achieved the highest test accuracy while also having a relatively short training time compared with the other tested models.

Comparison with Other Models

Three other architectures were also tested:

Model	Test Accuracy	Parameters	Train Time
CNN	100%	50,435	~2.9 s
SimpleRNN	100%	2,243	~4.1 s
CNN + LSTM	66.67%	9,603	~6.8 s
LSTM	64.67%	5,507	~6.5 s

The CNN was selected for the final activity because it provided excellent classification performance and fast training on this dataset.

Usage

The model can be loaded with TensorFlow/Keras:

import tensorflow as tf

model = tf.keras.models.load_model("activity_model.keras")