Gender Classification using Machine Learning

And testing it on various classifiers

The aim of the project is to choose the best gender classification model between -

• Logistic Regression
• Linear SVM
• Gaussian SVM
• Random Forest
• Adaptive Boost classifiers.

At the end we will also see how well each chosen model works.

Steps involved in the process:

1. Data Processing
2. Dimensionality Reduction
3. Fit the default Classifier
4. Tune Hyperparameters using Grid Search
5. Estimating the Best Parameters and the Best Scores
6. Plot Learning Curve

A. DATA PRE-PROCESSING -

1. Import the Dataset

• Import and separate data for both the genders.

2. Process the Data

• Convert to Gray
• Detect faces using HAAR Cascade
• Crop the face
• Resize Image

3. Creating DataFrame

• Label data and save as a dataframe

B. Working with the Classifier

1. Creating Pipeline for data

• Dimensionality reduction using PCA.
• Converting features to scaler

2. Fit the Classifier

• Fit the model with default parameters
• Check for the default accuracy

3. Tune Hyperparameters

• Run an exhaustive Grid Search for the best scores
• Check for the corresponding parameters

Dimensionality Reduced from 6400 features to only 300 components

Best Scores obtained from the subsequent classifiers:

Classifier	Best Score	Parameters Tuned
Logistic Regression	0.951912	C , n_components
Linear SVM	0.945475	C , n_components
Gaussian SVM	0.970087	C , gamma
Random Forest	0.914048	n_estimators , max_depth
AdaBoost with Decision tree	0.936766	n_estimators , learning _rate

Gaussian SVM clearly shows the highest accuracy

Analyzing the results by plotting the Learning Curve

To configure the perfect bias-variance trade-off for the specific algorithm

Conclusion -

Altogether, Logistic Regression does a decently good job in predicting the scores and has a relatively better Learning curve.