Machine Learning: Nearest Neighbor (NN) algorithm for stock transfers between warehouses

 

🏭 Scenario: Nearest Neighbor Algorithm for Warehouse Stock Transfer

Background

A company, GrapheneTech, distributes graphene sheets across 5 regional warehouses in Southeast Asia:

Warehouse	Location	Average Daily Demand (units)	Average Lead Time (days)	Current Inventory (units)
W1	Kuala Lumpur	95	2	200
W2	Penang	75	1	70
W3	Johor	90	2	110
W4	Singapore	120	3	50
W5	Bangkok	60	5	300

---

Problem

• Warehouse W4 (Singapore) is facing a stock shortage because of sudden demand increase.

• The replenishment lead time from the supplier is 7 days, which is too slow.

• The company wants to transfer stocks from another warehouse temporarily to maintain W4’s 95% service level.

But which warehouse should W4 get help from?

---

Step 1: Define the Features for Similarity

Each warehouse is described by quantitative features that influence stock movement feasibility:

Feature	Meaning
$D$	Average daily demand
$L$	Lead time from supplier
$S$	Current stock level
$Dist$	Transportation distance (km) from W4

Suppose distances from W4 are:

From → W4	Distance (km)
W1	350
W2	600
W3	250
W5	1400

---

Step 2: Represent Each Warehouse as a Feature Vector

Each warehouse = $[D, L, S, Dist]$

Warehouse	Vector $[D, L, S, Dist]$
W1	[95, 2, 200, 350]
W2	[75, 1, 70, 600]
W3	[90, 2, 110, 250]
W5	[60, 5, 300, 1400]
W4 (Target)	[120, 3, 50, 0]

---

Step 3: Apply Nearest Neighbor (Euclidean Distance)

We compute the “distance” between W4 and every other warehouse:

$$d(W_i, W4) = \sqrt{(D_i - D_4)^2 + (L_i - L_4)^2 + (S_i - S_4)^2 + (Dist_i - 0)^2}$$

Let’s calculate digit by digit for clarity:

---

For W1:​≈381.7

---

For W2:​≈602.0

---

For W3:​≈258.8

---

For W5:​≈1423.4

---

Step 4: Find the Nearest Neighbor

Warehouse	Distance to W4	Rank
W3 (Johor)	258.8	🥇 Nearest
W1	381.7	2
W2	602.0	3
W5	1423.4	4

✅ Nearest Neighbor = W3 (Johor)

---

Step 5: Decision and Action

Since W3 is most similar and geographically close:

• W3 can transfer 40 units of graphene sheets to W4 immediately.

• This transfer minimizes cost and maintains regional balance.

---

Step 6: Result

Warehouse	Inventory Before	Transfer	Inventory After
W3	110	−40	70
W4	50	+40	90

• W4’s shortage problem solved.

• W3 still has sufficient inventory for its demand.

---

🧠 Interpretation

The nearest neighbor algorithm here:

• Quantifies warehouse similarity (in demand, stock, lead time, and distance).

• Chooses the most appropriate source warehouse for emergency stock transfer.

• Avoids manual guesswork or arbitrary selection.

Python Code

import numpy as np

import pandas as pd

import matplotlib.pyplot as plt

# --- Step 1: Define warehouse data ---

data = {

    "Warehouse": ["W1", "W2", "W3", "W5"],   # W4 is the target

    "Demand": [95, 75, 90, 60],

    "LeadTime": [2, 1, 2, 5],

    "Stock": [200, 70, 110, 300],

    "DistanceToW4": [350, 600, 250, 1400]  # distance from each to W4

}

warehouses = pd.DataFrame(data)

# Target warehouse (W4)

target = np.array([120, 3, 50, 0])  # Demand, LeadTime, Stock, Distance

target_name = "W4"

# --- Step 2: Compute Euclidean distances ---

def euclidean_distance(row, target):

    return np.sqrt(((row - target) ** 2).sum())

features = warehouses[["Demand", "LeadTime", "Stock", "DistanceToW4"]].values

warehouses["NN_Distance"] = [euclidean_distance(row, target) for row in features]

# --- Step 3: Find nearest neighbor ---

nearest = warehouses.loc[warehouses["NN_Distance"].idxmin()]

print("Nearest warehouse to W4 (for stock transfer):")

print(nearest[["Warehouse", "NN_Distance"]])

# --- Step 4: Optional: simulate transfer ---

transfer_qty = 40

warehouses.loc[warehouses["Warehouse"] == nearest["Warehouse"], "Stock"] -= transfer_qty

w4_stock_after = 50 + transfer_qty

print("\nAfter transfer:")

print(f"{nearest['Warehouse']} new stock: {int(warehouses.loc[warehouses['Warehouse']==nearest['Warehouse'],'Stock'])}")

print(f"W4 new stock: {w4_stock_after}")

# --- Step 5: Visualization ---

plt.figure(figsize=(8,6))

plt.scatter(warehouses["DistanceToW4"], warehouses["Stock"], color="skyblue", s=100, label="Other Warehouses")

# Highlight nearest neighbor

plt.scatter(nearest["DistanceToW4"], nearest["Stock"], color="orange", s=200, edgecolors="black", label=f"Nearest: {nearest['Warehouse']}")

# Target warehouse (W4)

plt.scatter(0, 50, color="red", s=200, marker="*", label="Target: W4")

# Labels

for _, row in warehouses.iterrows():

    plt.text(row["DistanceToW4"]+10, row["Stock"]+5, row["Warehouse"], fontsize=10)

plt.text(10, 50+10, "W4", color="red", fontsize=11, weight="bold")

plt.title("Graphene Warehouse Stock vs Distance (Nearest Neighbor Transfer)")

plt.xlabel("Distance to W4 (km)")

plt.ylabel("Current Stock (units)")

plt.legend()

plt.grid(True, linestyle="--", alpha=0.6)

plt.tight_layout()

plt.show()