Sales Forecast project

This document provides a detailed workflow of the Iron Kaggle project, focusing on predicting future sales for a retail chain using historical sales data. The project follows a structured approach, incorporating data preprocessing, exploratory data analysis (EDA), feature engineering, machine learning modeling, and evaluation against actual sales. Specific libraries, file names, and steps are highlighted below.

Step 1: Importing Libraries and Reading Data

The project starts with importing necessary Python libraries:

• pandas and numpy for data manipulation,
• seaborn and matplotlib.pyplot for visualization,
• custom eda_functions.py for exploratory data analysis.

Data is read from two Excel files:

• Past sales data: Past_Data.xlsx, containing historical sales information.
• Future sales data: Future_data_without_label.xlsx, intended for making sales predictions.

import pandas as pd
import numpy as np
import seaborn as sns
import matplotlib.pyplot as plt
from eda_functions import standard_head, analyze_data

Step 2: Initial Data Analysis

Using pandas, the datasets are loaded and initially analyzed to understand their structure, using custom functions from eda_functions.py for summarizing and analyzing the data. This step is crucial for planning the preprocessing and modeling strategy.

Step 3: Data Cleaning and Preparation

Data cleaning involves creating new date-related columns (month, year, week, day) using pandas' datetime functionality and filtering out rows irrelevant for modeling (e.g., days when stores were closed). This step ensures the data is in the right format for analysis and modeling.

Step 4: Exploratory Data Analysis (EDA)

EDA is conducted to uncover sales trends and factors influencing sales, utilizing seaborn and matplotlib for visualization. Insights from this phase guide the feature engineering and modeling steps.

Step 5: Feature Engineering

New features are created to enrich the models, including interactions between customers and promotions, categorization of customer numbers, and calculation of average payments by customers. This process leverages numpy and pandas for data manipulation, enhancing the dataset with information believed to improve model predictions.

Step 6: Machine Learning Modeling

Several machine learning models are evaluated:

• Linear Regression, Decision Tree, Random Forest, and CatBoost from sklearn and catboost libraries.
• Models are trained on the historical data (Past_Data.xlsx) and evaluated using metrics like RMSE, MAE, and R-squared to select the best performer.

from sklearn.linear_model import LinearRegression
from sklearn.tree import DecisionTreeRegressor
from sklearn.ensemble import RandomForestRegressor
from catboost import CatBoostRegressor
from sklearn.metrics import mean_squared_error, mean_absolute_error, r2_score

Step 7: Preparing and Exporting Predictions

The final model predictions are prepared for the future dataset (Future_data_without_label.xlsx) and exported to an Excel file named predictions_4.xlsx. This step involves merging predictions with the future dataset template and ensuring the format aligns with project requirements.

Step 8: Evaluation Against Actual Sales

After appending the actual sales to the final_prediction_with_actual_metrics.xlsx file, the model's predictions are evaluated against these actual figures using:

• RMSE: 493
• MAE: 306
• R-squared (R2): 0.985

This evaluation confirms the model's accuracy and effectiveness in predicting future sales, showcasing the project's success in achieving its objective. The pandas library is utilized for data manipulation during this final evaluation phase, with metrics calculated using functions from the sklearn.metrics module.

Conclusion

The Sales Forecast project illustrates a comprehensive approach to sales forecasting, from data preprocessing and exploration through to predictive modeling and evaluation. By carefully selecting features, employing a variety of modeling techniques, and rigorously assessing performance, the project achieves a high degree of accuracy in predicting future sales, demonstrating the potential of data science in retail sales forecasting.