ninawekunal / applied-ml-algorithms

A ML project that takes the titanic dataset and plays with different ML models in scikit-learn and hyper-parameter tuning.

Applied-ML-Algorithms

Link: Titanic Prediction 5 models: (Open in new tab)

Objective:

• Take dataset from Kaggle and play with various Machine Learning models to get the best model which has highest accuracy in prediction.
• Preprocess the dataset by explore and cleaning it.
• Scenarios on when to use each model.
• Learn practical application of Machine Learning models like:
  • Linear Regression
  • Logistic Regression
  • Support Vector Machines(SVM)
  • Random Forest Classifiers
  • XG Boosting
  • Multi-Layer Perceptron
• Fine tune hyper-parameters of each model to improve accuracy:
  • Learning Rate
  • Depth of Decision Tree
  • num_estimators of decision tree
  • Regularization parameter C
  • Kernel
  • activation functions
  • hidden_layer_size
  • and much more...
• Choosing optimal model by testing each model on test dataset and evaulating against each other's score.

Result:

• After training 5 different models on the titanic dataset's validation data, I got an performance metrics as follows:
  • LR -- Accuracy: 0.758 / Precision: 0.778 / Recall: 0.675 / Latency: 2.2ms
  • SVM -- Accuracy: 0.753 / Precision: 0.767 / Recall: 0.675 / Latency: 4.3ms
  • MLP -- Accuracy: 0.742 / Precision: 0.776 / Recall: 0.627 / Latency: 11.5ms
  • RF -- Accuracy: 0.787 / Precision: 0.846 / Recall: 0.663 / Latency: 9.5ms
  • XGB -- Accuracy: 0.798 / Precision: 0.862 / Recall: 0.675 / Latency: 6.9ms
• XGB performs the best in terms of Accuracy of 79.8% and Precision of 86.2%, and ties with LR with a high recall of 67.5% with a low latency.
• On the test data, metrics were:
  • XGB -- Accuracy: 0.832 / Precision: 0.792 / Recall: 0.655 / Latency: 4.5ms
  • High Accuracy with a slightly low precision and recall with a low latency.