Machine Learning
Capstone Projects
Portfolio

from sklearn.ensemble import RandomForestRegressor from xgboost import XGBRegressor from sklearn.metrics import mean_squared_error import numpy as np, pandas as pd # Time-based split — NO data leakage split = df['date'].max() - pd.DateOffset(months=6) train, test = df[df.date <= split], df[df.date > split] FEATURES = ['day_of_week','month','quarter','year', 'week_of_year','is_weekend','store_id', 'item_id','price','kcal','store_avg_sales',...] models = { 'Random Forest' : RandomForestRegressor(n_estimators=200, random_state=42), 'XGBoost' : XGBRegressor(n_estimators=300, learning_rate=0.05, max_depth=6, random_state=42) } for name, m in models.items(): m.fit(train[FEATURES], train['item_count']) preds = m.predict(test[FEATURES]) rmse = np.sqrt(mean_squared_error(test['item_count'], preds)) print(f"{name}: RMSE = {rmse:.4f}") # Best model → 365-day forward forecast future = pd.date_range(df.date.max() + pd.Timedelta('1D'), periods=365) # ... generate feature rows for each future date × store × item ... df_forecast['predicted_count'] = np.maximum(0, best_model.predict(X_future))

import tensorflow as tf from tensorflow.keras.applications import EfficientNetB0 from tensorflow.keras import layers, Model # Load pre-trained backbone — FROZEN base = EfficientNetB0(include_top=False, weights='imagenet', input_shape=(224,224,3)) base.trainable = False # Build custom 11-class head x = layers.GlobalAveragePooling2D()(base.output) x = layers.Dense(512, activation='relu')(x) x = layers.Dropout(0.4)(x) x = layers.Dense(256, activation='relu')(x) x = layers.Dropout(0.3)(x) out = layers.Dense(11, activation='softmax')(x) model = Model(base.input, out) model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy']) # Custom callback: stop at val_accuracy >= 90% class StopAtAccuracy(tf.keras.callbacks.Callback): def on_epoch_end(self, epoch, logs=None): if logs.get('val_accuracy', 0) >= 0.90: self.model.stop_training = True # Train WITH augmentation from tensorflow.keras.preprocessing.image import ImageDataGenerator aug = ImageDataGenerator(rescale=1/255, rotation_range=20, horizontal_flip=True, zoom_range=0.2, width_shift_range=0.2) history = model.fit(aug.flow_from_directory(TRAIN_DIR, target_size=(224,224), batch_size=32, class_mode='categorical'), validation_data=test_gen, epochs=50, callbacks=[StopAtAccuracy()])

from surprise import SVD, Dataset, Reader, accuracy from surprise.model_selection import cross_validate, train_test_split # Build Surprise dataset from ratings CSV reader = Reader(rating_scale=(1, 5)) data = Dataset.load_from_df( ratings[['User_Id', 'Place_Id', 'Place_Ratings']], reader) trainset, testset = train_test_split(data, test_size=0.2, random_state=42) # SVD — matrix factorisation with 50 latent factors svd = SVD(n_factors=50, n_epochs=30, lr_all=0.005, reg_all=0.02) svd.fit(trainset) preds = svd.test(testset) print(f"RMSE: {accuracy.rmse(preds):.4f}") # Recommend top-5 unvisited places for any user def recommend_places(user_id, n=5): visited = ratings[ratings.User_Id == user_id].Place_Id.tolist() candidates = [p for p in places.Place_Id.unique() if p not in visited] preds_list = [(p, svd.predict(user_id, p).est) for p in candidates] top_n = sorted(preds_list, key=lambda x: x[1], reverse=True)[:n] return places[places.Place_Id.isin([p[0] for p in top_n])] recommend_places(user_id=1, n=5)