Editing Ai Supply Chain (section)

== <span style="color: #FFFFFF;">Applying</span> ==
'''Demand forecasting with LightGBM:'''
<syntaxhighlight lang="python">
import pandas as pd
import lightgbm as lgb
from sklearn.metrics import mean_absolute_percentage_error
import numpy as np

# Load sales data: item, store, date, sales, price, promotions, weather
df = pd.read_csv("sales_data.csv", parse_dates=['date'])
df = df.sort_values(['item_id', 'store_id', 'date'])

# Feature engineering
df['day_of_week'] = df['date'].dt.dayofweek
df['month'] = df['date'].dt.month
df['week_of_year'] = df['date'].dt.isocalendar().week.astype(int)
# Lag features (historical sales)
for lag in [7, 14, 28, 56]:
    df[f'sales_lag_{lag}'] = df.groupby(['item_id','store_id'])['sales'].shift(lag)
# Rolling mean features
for window in [7, 28]:
    df[f'sales_rolling_mean_{window}'] = (
        df.groupby(['item_id','store_id'])['sales']
        .shift(1).rolling(window).mean().values
    )

features = ['item_id','store_id','day_of_week','month','week_of_year',
            'price','is_promotion'] + [c for c in df.columns if 'lag' in c or 'rolling' in c]
df = df.dropna()

# Chronological train/test split
cutoff = df['date'].max() - pd.Timedelta(days=28)
train = df[df['date'] <= cutoff]
test = df[df['date'] > cutoff]

model = lgb.LGBMRegressor(n_estimators=500, learning_rate=0.05, num_leaves=127)
model.fit(train[features], train['sales'],
          eval_set=[(test[features], test['sales'])],
          callbacks=[lgb.early_stopping(50)])

preds = model.predict(test[features])
print(f"MAPE: {mean_absolute_percentage_error(test['sales'], preds):.2%}")
</syntaxhighlight>

; Supply chain AI tools
: '''Demand forecasting''' → LightGBM, TFT, Amazon Forecast, Google Cloud Demand Forecasting
: '''Route optimization''' → Google OR-Tools, OptaPlanner, Routific; ML-enhanced with pointer networks
: '''Warehouse management''' → Locus Robotics, Geek+, Symbotic (robot + AI)
: '''Visibility platforms''' → Project44, FourKites — real-time tracking + ETA prediction
: '''Risk monitoring''' → Resilinc, Everstream, Riskmethods — NLP on news/supplier signals
</div>

<div style="background-color: #8B4500; color: #FFFFFF; padding: 20px; border-radius: 8px; margin-bottom: 15px;">