KNN/KNNModel.py

import os
import pandas as pd
import numpy as np
from sklearn.preprocessing import LabelEncoder, StandardScaler
from sklearn.model_selection import train_test_split, cross_val_score
from sklearn.neighbors import KNeighborsRegressor
from sklearn.metrics import r2_score, mean_squared_error
import matplotlib.pyplot as plt

class KNNModel:
    def __init__(self):
        self.target_column = "Fat_Percentage"
        self.data_path = os.path.join("data", "gym_members_exercise_tracking.csv")  # Updated path
        self.data = None
        self.selected_features = None
        self.label_encoders = {}
        self.scaler = StandardScaler()
        self.knn = None
        self.optimal_k = None

    def load_and_preprocess_data(self):
        # Load data
        self.data = pd.read_csv(self.data_path)

        # Check if 'Experience_Level' column exists
        if 'Experience_Level' not in self.data.columns:
            raise ValueError("'Experience_Level' column not found in the dataset.")

        # Encode categorical features
        categorical_features = ['Gender', 'Experience_Level', 'Workout_Type']
        for feature in categorical_features:
            if feature in self.data.columns:  # Ensure the column exists
                le = LabelEncoder()
                self.data[feature] = le.fit_transform(self.data[feature])
                self.label_encoders[feature] = le  # Save the encoder
            else:
                raise ValueError(f"'{feature}' column not found in the dataset.")

        # Compute correlations and select features
        correlation_matrix = self.data.corr()
        target_corr = correlation_matrix[self.target_column].sort_values(ascending=False)
        self.selected_features = target_corr[abs(target_corr) >= 0.5].index.tolist()
        self.selected_features.remove(self.target_column)

        # Prepare dataset
        X = self.data[self.selected_features]
        y = self.data[self.target_column]
        X_scaled = self.scaler.fit_transform(X)

        # Split into training and testing sets
        return train_test_split(X_scaled, y, test_size=0.2, random_state=42)

    def find_optimal_k(self, X_train, y_train, k_range=20):
        k_values = range(1, k_range + 1)
        cv_scores = []

        for k in k_values:
            knn = KNeighborsRegressor(n_neighbors=k)
            scores = cross_val_score(knn, X_train, y_train, cv=5, scoring='r2')
            cv_scores.append(scores.mean())

        # Find optimal k
        self.optimal_k = k_values[np.argmax(cv_scores)]

        # Plot k vs. R^2
        plt.figure(figsize=(10, 6))
        plt.plot(k_values, cv_scores, marker='o', linestyle='-')
        plt.xlabel("Number of Neighbors (k)")
        plt.ylabel("Cross-Validated R^2")
        plt.title("R^2 vs. k")
        plt.grid()
        plt.savefig(os.path.join("KNN", "optimal_k_plot.png"))
        plt.close()

        return self.optimal_k

    def train_model(self, X_train, y_train):
        if not self.optimal_k:
            raise ValueError("Optimal k is not set. Run find_optimal_k() first.")
        self.knn = KNeighborsRegressor(n_neighbors=self.optimal_k)
        self.knn.fit(X_train, y_train)

    def evaluate_model(self, X_test, y_test):
        if not self.knn:
            raise ValueError("Model is not trained. Run train_model() first.")
        y_pred = self.knn.predict(X_test)
        r2 = r2_score(y_test, y_pred)
        mse = mean_squared_error(y_test, y_pred)
        return r2, mse

    def predict(self, input_data: pd.DataFrame) -> float:
        if not self.knn:
            raise ValueError("Model is not trained. Run train_model() first.")
        input_scaled = self.scaler.transform(input_data)
        return self.knn.predict(input_scaled)[0]