Mastering E-Commerce Trend Forecasting: A Practical Guide to Data Extraction and Machine Learning with Lazada API

2025-04-18

1. Introduction: Why Trend Forecasting Matters in E-Commerce

As Southeast Asia’s e-commerce market continues to grow rapidly, competition among sellers is becoming increasingly intense. The ability to accurately forecast product trends and shifts in consumer demand has become a critical advantage for merchants looking to optimize operations and maximize profitability.

Traditional experience-based decision-making is no longer sufficient in a data-driven world. By leveraging API data extraction and machine learning models, businesses can build automated, intelligent forecasting systems that detect emerging patterns and help them stay ahead of the curve.

Lazada, one of the leading e-commerce platforms in Southeast Asia, offers rich datasets through its API — including order, product, and inventory data — making it an ideal foundation for market trend analysis.

2. Extracting Historical Sales Data via Lazada API

2.1 Overview of Available APIs

Lazada’s official Lazada API provides access to key data endpoints, including:

GetOrders: Retrieve order summaries
GetOrderItems: Access detailed order items
GetProducts: Fetch product listings
GetCategoryTree: Browse category structure

Third-party platforms such as Luckdata also offer simplified endpoints for quicker access:

get lazada product detail: Retrieve product details
search lazada product list: Query product search results

2.2 Python Implementation Example

Here’s a basic implementation of Lazada API using Python:

import requests
import pandas as pd
access_token = "YOUR_ACCESS_TOKEN"
url = "https://api.lazada.com.my/orders/get"
params = {
"access_token": access_token,
"created_after": "2024-01-01T00:00:00+08:00",
"limit": 100,
"offset": 0
}
response = requests.get(url, params=params)
orders = response.json()["data"]["orders"]
# Save data to local CSV
df = pd.DataFrame(orders)
df.to_csv("orders_2024.csv", index=False)

Using Luckdata for product lookup:

import requests
headers = {
'X-Luckdata-Api-Key': 'your_key'
}
response = requests.get(
'https://luckdata.io/api/lazada-online-api/x3fmgkg9arn3?site=vn&itemId=2396338609',
headers=headers
)
print(response.json())

3. Data Cleaning and Preprocessing Workflow

3.1 Common Issues to Address

After retrieving the data, common issues typically include:

Missing values (e.g., price or quantity is null)
Outliers (e.g., extremely high or low sales values)
Data type inconsistencies (e.g., dates stored as strings)

3.2 Cleaning Example with Pandas

df = pd.read_csv("orders_2024.csv")
# Remove canceled orders
df = df[df['status'] != 'canceled']
# Fill missing prices with the mean
df['price'] = df['price'].fillna(df['price'].mean())
# Convert timestamps
df['created_at'] = pd.to_datetime(df['created_at'])
# Retain necessary columns
df = df[['created_at', 'product_id', 'quantity', 'price']]

✅ Tip: The quality of your preprocessing has a direct impact on model accuracy — always double-check dates, prices, and status labels.

4. Feature Engineering: Preparing the Data for Forecasting Models

4.1 Time Series Feature Construction

df['date'] = df['created_at'].dt.date
daily_sales = df.groupby(['date', 'product_id'])['quantity'].sum().reset_index()
# Create time series format
pivot_sales = daily_sales.pivot(index='date', columns='product_id', values='quantity').fillna(0)

4.2 Adding Cyclical and Holiday Features

pivot_sales['day_of_week'] = pd.to_datetime(pivot_sales.index).dayofweek
pivot_sales['is_weekend'] = pivot_sales['day_of_week'].isin([5,6]).astype(int)

✅ Bonus: You can also add custom flags for holidays, promotional campaigns (e.g., 11.11), and seasonal sales events.

5. Forecasting Model Selection and Implementation

5.1 ARIMA for Individual Product Forecasting

from statsmodels.tsa.arima.model import ARIMA
import matplotlib.pyplot as plt
ts = pivot_sales[12345]  # Example product ID
ts = ts.asfreq('D').fillna(0)
model = ARIMA(ts, order=(1,1,1))
model_fit = model.fit()
forecast = model_fit.forecast(steps=30)
ts.plot(label='Historical Sales')
forecast.plot(label='Forecasted Sales', color='red')
plt.legend()
plt.title("Product 12345 Sales Forecast")
plt.show()

5.2 LSTM Deep Learning Model for Multiple Products

from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import LSTM, Dense
from sklearn.preprocessing import MinMaxScaler
import numpy as np
ts = pivot_sales[12345].values.reshape(-1,1)
scaler = MinMaxScaler()
ts_scaled = scaler.fit_transform(ts)
def create_dataset(data, step=7):
X, y = [], []
for i in range(len(data)-step):
X.append(data[i:i+step])
y.append(data[i+step])
return np.array(X), np.array(y)
X, y = create_dataset(ts_scaled)
model = Sequential()
model.add(LSTM(50, input_shape=(X.shape[1], X.shape[2])))
model.add(Dense(1))
model.compile(loss='mse', optimizer='adam')
model.fit(X, y, epochs=20, batch_size=16, verbose=1)

Model Comparison:

ARIMA: Best for single-product, linear trends
LSTM: Captures nonlinear patterns and seasonal variations across multiple SKUs
Prophet or Transformer: Suitable for long-term or multi-seasonal forecasting

6. Model Evaluation and Visualization

6.1 Forecast Accuracy with RMSE

from sklearn.metrics import mean_squared_error
pred = model.predict(X)
rmse = np.sqrt(mean_squared_error(y, pred))
print(f"LSTM Forecast RMSE: {rmse:.4f}")

6.2 Visualizing Predictions vs Actual Sales

plt.plot(scaler.inverse_transform(y.reshape(-1,1)), label='Actual')
plt.plot(scaler.inverse_transform(pred), label='Predicted', linestyle='--')
plt.legend()
plt.title("LSTM Forecast Performance")
plt.show()

✅ Tip: You can also evaluate MAE, MAPE, or use cross-validation to verify robustness.

7. Real-World Applications of Forecasting Models

Inventory Optimization: Automate restocking based on forecasted demand
Campaign Planning: Identify peak sales periods and align promotions accordingly
Business Intelligence Integration: Visualize trends using BI tools like Power BI or Tableau
Supply Chain Strategy: Detect high-performing vs underperforming SKUs early

8. Advanced Use Cases and System Expansion

Use Facebook Prophet for handling multiple seasonal patterns and holidays
Apply Transformer models for enhanced long-term forecasting
Integrate with Kafka / Airflow for real-time forecasting and scheduling
Build a prediction API to serve forecasts to internal systems or apps

9. Conclusion: Empower Your Business with Predictive Insights

This end-to-end guide walks you through the full lifecycle — from data extraction and cleaning to modeling and deployment — for building a scalable, intelligent forecasting solution based on Lazada data.

Whether you're a data analyst, an e-commerce operations manager, or a software engineer, this pipeline empowers you to make faster, smarter, and data-driven decisions.

In the future of commerce, decision-making will no longer rely on guesswork — predictive intelligence will be the new competitive edge.