2.2  Association Rule Learning

import numpy as np
import pandas as pd
from mlxtend.frequent_patterns import apriori, association_rules
from mlxtend.preprocessing import TransactionEncoder

np.random.seed(42)

# ── Simulate Bank Product Holdings ────────────────────────────
products = ['Savings','Debit','Credit','Mortgage','Insurance','Investment','Loan']

def simulate_basket(n_customers):
    baskets = []
    for _ in range(n_customers):
        b = []
        if np.random.rand() < 0.80: b.append('Savings')
        if np.random.rand() < 0.70 and 'Savings' in b: b.append('Debit')
        elif np.random.rand() < 0.30: b.append('Debit')
        if np.random.rand() < 0.45: b.append('Credit')
        if np.random.rand() < 0.25: b.append('Mortgage')
        if 'Mortgage' in b and np.random.rand() < 0.70: b.append('Insurance')
        if np.random.rand() < 0.20 and ('Savings' in b or 'Credit' in b): b.append('Investment')
        if np.random.rand() < 0.15: b.append('Loan')
        if b: baskets.append(b)
    return baskets

transactions = simulate_basket(3000)

# ── One-Hot Encode ────────────────────────────────────────────
te = TransactionEncoder()
te_array = te.fit_transform(transactions)
df = pd.DataFrame(te_array, columns=te.columns_)

# ── Frequent Itemsets ─────────────────────────────────────────
frequent_items = apriori(df, min_support=0.10, use_colnames=True)
frequent_items['length'] = frequent_items['itemsets'].apply(len)
print(f"=== Apriori — Bank Products ===")
print(f"Frequent itemsets: {len(frequent_items)}")
print(frequent_items.sort_values('support', ascending=False).head(10).to_string(index=False))

# ── Association Rules ─────────────────────────────────────────
rules = association_rules(frequent_items, metric='confidence', min_threshold=0.5)
rules = rules.sort_values('lift', ascending=False)
print(f"\nTop 10 Rules by Lift:")
print(rules[['antecedents','consequents','support','confidence','lift']].head(10).to_string(index=False))

# ── Strong rules (lift > 1.5) ─────────────────────────────────
strong = rules[rules['lift'] > 1.5]
print(f"\nStrong rules (lift>1.5): {len(strong)}")

import numpy as np
import pandas as pd
import time
from mlxtend.frequent_patterns import fpgrowth, apriori, association_rules
from mlxtend.preprocessing import TransactionEncoder

np.random.seed(42)

# ── Simulate Symptom-Disease Records ─────────────────────────
symptoms = ['Fever','Cough','Fatigue','Headache','Nausea','SoreThroat',
            'Dyspnoea','ChestPain','Diarrhoea','Rash']

def gen_record(profile):
    base, extras = profile
    items = list(base) + [s for s in extras if np.random.rand() < 0.4]
    return list(set(items))

profiles = [
    (['Fever','Cough'], ['Fatigue','SoreThroat','Headache']) for _ in range(4000)
] + [
    (['ChestPain','Dyspnoea'], ['Fatigue','Nausea']) for _ in range(1200)
] + [
    (['Diarrhoea','Nausea'], ['Fatigue','Fever']) for _ in range(1500)
] + [
    ([s],['Headache','Rash']) for s in np.random.choice(symptoms, 3300)
]

transactions = [gen_record(p) for p in profiles]
valid = [t for t in transactions if len(t) >= 1]

te = TransactionEncoder()
df = pd.DataFrame(te.fit_transform(valid), columns=te.columns_)

# ── FP-Growth vs Apriori timing ───────────────────────────────
min_sup = 0.05
t0 = time.time()
freq_fp = fpgrowth(df, min_support=min_sup, use_colnames=True)
t_fp = time.time() - t0

t0 = time.time()
freq_ap = apriori(df, min_support=min_sup, use_colnames=True)
t_ap = time.time() - t0

print("=== FP-Growth vs Apriori ===")
print(f"FP-Growth : {len(freq_fp):4d} itemsets in {t_fp:.4f}s")
print(f"Apriori   : {len(freq_ap):4d} itemsets in {t_ap:.4f}s")
print(f"Speedup   : {t_ap/max(t_fp,1e-6):.2f}x")

# ── Association Rules from FP-Growth ─────────────────────────
rules = association_rules(freq_fp, metric='confidence', min_threshold=0.6)
rules = rules.sort_values('lift', ascending=False)
print("\nTop 8 Rules (Symptom-Disease Patterns):")
print(rules[['antecedents','consequents','support','confidence','lift']].head(8).to_string(index=False))