Careers

Data Science Forecast Assignment

Overview

You have been given a dataset that tracks weekly sales quantities (the "quantity" column) for a specific product (or set of products) defined by fields like weekend_date, channel, brand, category, sub_category, and SerialNum .

Your task is to:

• Explore and understand the data (EDA).
• Build a forecasting model to predict weekly sales for the next 3 months (September, October, November).
• Report the Validation Accuracy for Last-3-months in the data i.e. Jun-Jul-Aug.
• Provide your forecast results and measure monthly accuracy.
• Share your work in two Jupyter Notebooks (one for EDA and one for modeling) and include final model objects for reproducibility.

Data Description

• weekend_date: The date marking the end of each week (weekly time index).
• channel: Sales channel identifier.
• brand: Product brand identifier.
• category & sub_category: Product classification details.
• SerialNum: Unique product identifier.
• quantity: Weekly quantity sold (target variable for forecasting).

These columns provide a clear view of weekly sales history for a product. You will focus primarily on weekend_date (as your time variable) and quantity (as your numeric series to forecast).

Assignment Tasks

1. Exploratory Data Analysis (EDA) Notebook

• Data Inspection:
  • Load and inspect the dataset (e.g., checking for missing values, outliers, or data inconsistencies).
• Visualizations:
  • Plot weekly sales over time, look for trends or seasonal patterns, and check if there are any anomalies.
• Initial Insights:
  • Summarize key findings that might influence your forecasting approach (e.g., seasonality around certain months).

2. Forecasting & Modeling Notebook

• Data Preparation:
  • Confirm the data is at weekly intervals without gaps (or address missing weeks as needed).
  • Split your data into training (historical) and validation/future periods.
• Model Building:
  • Select at least one machine-learning approach.
  • Fit your model(s) on the historical weekly data.
  • Save the fitted model object(s) (e.g., as .pkl or .joblib files) for reproducibility.
• Validation:
  • Perform Validation on Last 3 months i.e. whole of Jun-Jul-Aug 2024.
  • Report the monthly Accuracy for the same.
• Forecast Generation:
  • Produce weekly forecasts for the next 3 months (i.e., September, October, November).
  • Present a clear table or DataFrame with week-ending dates and predicted quantities.

3. Monthly Accuracy Evaluation

Once actual data is available (or if you perform a backtest), calculate accuracy using:

Where:

• Σ represents summation.
• ŷ is the predicted target.
• y is the actual target.

Report this value for each month Jun-Jul-Aug 2024.
Provide commentary on how the model performed and any observed trends or discrepancies.

Deliverables

• EDA Notebook:
  • Exploration of the data (descriptive statistics, time-series plots, insights).
  • Any data-cleaning or transformation steps you performed..
• Modeling Notebook:
  • Explanation of the model selection and fitting process (code + rationale)..
  • Weekly forecast results for Sept–Nov for every serial number(e.g., a DataFrame with weekend_date and forecasted quantity).
  • Monthly accuracy calculations (if you have actuals for that time period or through a backtest approach).
  • Model Objects: Include the serialized/fitted model file(s) so that anyone can load the model and replicate your forecasts without re-fitting from scratch.
• Presentation/Discussion:
  • Summary of your findings in both notebooks, focusing on data insights, forecast performance, and recommended next steps for improving or extending the model.

Notes & Suggestions

• You can either create a separate model for each time series or use a more comprehensive technique that handles multiple series together.
• Take care to document your assumptions (e.g., handling missing data or outliers)..
• For reproducibility, ensure you version-control your code and clearly state any library dependencies (e.g., pandas, statsmodels, scikit-learn, prophet, etc.).

Good luck! 🚀