Vision & Hand State Detection

This project came to summarize the concepts we learn over this semester.

• Unsupervised Machine learning models
• Model Based on Boosting and Bagging technique
• Dimensionality reduction technique
• Pandas version 1.0

Part A:

In this part we take the Fashion Mnist dataset and use dimensionality reduction techniques like PCA and SVD

to extract the most importance features(from 784 to 23).

After that we use model based boosting and other basic models.

Also, we use in unsupervised model like k means and "translate" the clustering to our labels(types of clothes)

Note: As A bonus we use conventional neural network

Scores Summery:

Using machine learning regular model:

model name	score without PCA	score with PCA
Naive Bayes(Gaussian)	58%	74%
K Nearest Neighbors	86%
Decision Tree	81%	76%
SVC	87%	89%
K Means(unsupervised)	78%	78%

Using ensemble learning:

model name	score without PCA	score with PCA
Random Forest	88%	86%
Adaptive Boosting (adaboost)	82%	78%
XgBoost	90%

Using deep learning:

model name	score
neural network(Sequential)	84%
Sequential with layers change	90%

Part B:

In this part we have dataset with hands motion sensor recording we our task is to classify the type of the hands moving:

1. Spontaneous
2. Synchronized
3. Alone

The features we get from the sensor is:

• Time
• Frame ID
• Hand Type
• number of hands
• Position(X,Y,Z)
• Velocity(X,Y,Z)
• Pitch
• Roll
• Yaw
• Wrist Position(X,Y,Z)
• Elbow Position(X,Y,Z)
• Grab Strength
• Grab Angle
• Pinch Strength

Note: In very record we ignore the first 7 seconds. Also, we combine every 5 rows into 1.

The Results

Model Name	Validation Score
K Nearest Neighbors	88%
Sequential Neural Net	79%
XgBoost	100%
Adaptive Boosting	99%
Random Forest	95%
Decision Tree	94%

Sources

• Intro to data science - washington university - svd part
• Intro to Machine Learning - fastai - parts
• Relationship between svd and pca
• Why xgboost so good
• Ensemble methods
• ML approaches for time series
• Git
• Fast.ai course - Practical Deep Learning for Coders-just the begin
• Udacity Course
• Gradient Descent

and other from medium,towards data science...

By: Yosef Danan

Really Cool Project!