Machine Learning - Exploration

This repo is an ongoing exploration of Machine Learning models.

Current objectives:

• Unsupervised models
• Hyperparameter Tuning

Multi-Linear Regression

Not the best model because of its low r-score: 0.104...

The multi_linear regression is better suited for "binary problems".

The dataset we have is more aligned with a classification model.

KNN

The KNN model produce a moderate, initial r-score: 0.832...

Parameters were adjusted by importing GridSearchCV. These parameters were:

• n_neighbors
• weights
• algorithm

After parameter adjustment, there was a small increase in r-score: 0.837...

best given parameters: {'algorithm': 'auto', 'n_neighbors': 11, 'weights': 'distance'} 

Keras

Accuracy below 0.84 prior to adjusting 'number_hidden_nodes' parameter.

Adjusted 'number_hidden_nodes' from 4 to 100, accuracy increased to 0.88 with a loss of 1.11.

from tensorflow.keras.layers import Dense
number_inputs = 40
number_hidden_nodes = 100
keras.add(Dense(units=number_hidden_nodes,
                activation='relu', input_dim=number_inputs))

Conclusion

The keras model seems to be the best performing model in terms of accuracy.

---

Contact

LinkedIn | https://www.linkedin.com/in/niko-elvambuena/
Email | niko.elvambuena95@gmail.com