Project Summary: In the "Kyphosis disease prediction and Imbalanced Data Handling" project, the objective was to develop a predictive model to identify the presence or absence of kyphosis, a medical condition. The dataset used in the project, named "Kyphosis.csv," contained imbalanced data, making it challenging to build an accurate model. The project's workflow can be summarized as follows:
1. Data Preprocessing:
--> The "Kyphosis.csv" dataset was loaded, which initially contained imbalanced data.
--> To address the class imbalance, the Synthetic Minority Over-sampling Technique (SMOTE) was applied to oversample the minority class, creating a balanced dataset.
2. Label Encoding:
--> As the target class was originally in string format ("sample"), label encoding was used to convert it into a numeric form.
3. Feature Scaling:
--> StandardScaler from scikit-learn was applied to scale the features, ensuring they had a mean of 0 and a standard deviation of 1.
4. Training-Testing Split:
--> The dataset was split into training and testing sets, enabling model evaluation.
5. Model Selection:
--> Two different models were considered for classification:
----> A Fully Connected Neural Networks (FCNNs) model built from scratch using PyTorch.
----> An XGBoost model with hyperparameter tuning using GridSearchCV.
6. Model Training and Evaluation:
--> The Simple Neural Network model was trained using the Adam optimizer and binary cross-entropy loss for binary classification. Training continued for 100 iterations.
--> The XGBoost model's hyperparameters were optimized using GridSearchCV to identify the best parameter combinations for the Kyphosis dataset.
--> After training both models, they were evaluated on the testing data.
--> Evaluation metrics, such as accuracy, precision, recall, f1 score, classification report, were used to assess the model's performance.
7. Model Comparison:
--> The project concluded with the finding that the XGBoost model outperformed the Fully Connected Neural Network model, achieving an accuracy of 0.96. In contrast, the FCNN model achieved an accuracy of 0.85.
The project aimed to demonstrate the process of handling imbalanced data, performing binary classification using machine learning (XGBoost with GridSearchCV) and deep learning (Fully Connected Neural Network) models, and selecting the best-performing model for predicting kyphosis in a medical context.