| You can download the initial exercise file from [link](https://quera.org/contest/assignments/103149/download_problem_initial_project/356093/). | | --- | Welcome to the Data Analysis and Strategy Department of **"Rockstar"**! The creator of one of the biggest and most profitable games in history: **Grand Theft Auto V**. More than a decade after the game's release, our economy and players' behavior have undergone profound changes. Now, **Rockstar** needs to predict its sales and profits to maximize sales of its new game version, but it has encountered a serious problem and has hired you to solve these issues. Due to a flaw in the legacy servers, the data related to **gross revenue** in 60 percent of the records from past years has been deleted. Worse yet, for the coming months, we don't know what percentage of players will purchase the game **digitally** and what percentage **physically** in order to adjust distribution and marketing budgets. You have joined the team to solve this crisis. The marketing manager wants a comprehensive system from you: a system that can examine past behavior to simultaneously predict which **sales channel (Physical or Digital)** a sales record belongs to and what the **exact revenue** generated from it will be.  The data engineering team has provided you with the extracted information in the form of two files, `train.csv` and `test.csv`. Note that in the training file, 60% of the revenue values are missing, and in the test file, both target columns have been completely removed for you to predict them. The details of the columns in this dataset are as follows: | **Column Name** | **Description** | | --- | --- | | `transaction_id` | Unique identifier for each record in the dataset | | `year` | Calendar year of the record | | `month` | Calendar month of the record (1 to 12) | | `quarter` | Financial quarter derived from the month (1 to 4) | | `country` | Name of the country where the sale was made | | `iso3_code` | Three-letter ISO code (ISO 3166-1 alpha-3) for the country | | `region` | Continent or geographical region of the country | | `platform` | Gaming platform (e.g., PS3, PS4, PC, Xbox Series X|S) | | `game_edition` | Purchased edition of the game (Standard, Premium, Legacy, etc.) | | `units_sold` | Number of copies sold in that month/country/platform | | `new_customers` | Number of buyers who purchased for the first time in that period | | `returning_customers` | Number of returning buyers (those who have purchased before) | | `estimated_active_players` | Estimated number of unique active players | | `peak_concurrent_players` | Highest number of concurrent players recorded in that month | | `online_players` | Number of active players in any of the online sections | | `story_mode_players` | Number of players engaged with the story mode content (single-player) | | `gta_online_players` | Number of active players specifically in the GTA Online section | | `average_playtime_hours` | Average playtime hours per active player in that month | | `average_session_length_minutes` | Average length of each gaming session in minutes | | `holiday_season` | Was the sale during the holiday season (November, December, January)? (0 or 1) | | `major_sale_event` | Name of the special sale event active in that month (if any) | | `marketing_campaign` | Was a marketing campaign active during that period? (0 or 1) | | `customer_rating` | Average rating submitted by users (from 4.2 to 5.0) | | `review_count` | Number of submitted reviews | | `refund_rate_percentage` | Percentage of refunded purchases | | `currency` | Local currency code of the target country | | `exchange_rate_to_usd` | Local currency to USD exchange rate at the time of sale | | `internet_penetration_percentage` | Internet penetration percentage in the target country | | `gaming_market_size` | Estimated relative size of the gaming market in the country | | `population_millions` | Population of the target country in millions | | `gdp_per_capita_usd` | GDP per capita of the country in USD (indicating purchasing power) | | `release_phase` | Platform release life cycle phase (Launch, Growth, Mature, Legacy) | | `weekend_sales_percentage` | Percentage share of sales on weekends | | `weekday_sales_percentage` | Percentage share of sales on weekdays | | `season` | Meteorological season (Winter, Spring, Summer, Autumn) | | `special_event` | Special event that stimulated sales (e.g., release of a major update) | | `top_game_category` | Main genre category of the game (Action-Adventure) | | `platform_generation` | Gaming console generation (e.g., Gen7, Gen8, Gen9, PC) | | `gross_revenue_usd` | **(First Target Column - Regression):** Gross revenue earned in USD | | `sales_channel` | **(Second Target Column - Classification):** Sales channel, with values `Physical` or `Digital` | ### **Predictions Output Format** Your model must make its predictions for all rows present in the test file. In the initial notebook file, code has been provided that saves your model's output in the format of a file named `submission.csv`. This file must include the following three columns: * `transaction_id`: Transaction ID * `sales_channel_pred`: Your prediction for the sales channel (`Physical` or `Digital`) * `revenue_pred`: Your prediction for the revenue amount in USD (a decimal number) ## How to Submit Your Answer After running the result generator cell (the last cell of the notebook file), submit the generated `result.zip` file. This zip file will automatically include your notebook and the `submission.csv` file. Please note that you must save the changes made in the notebook using the `ctrl+s` shortcut before running the result generator cell; otherwise, your **score** will be changed to **zero** at the end of the competition. Also, if you are using Colab to run this notebook file, download the latest version of your notebook and place it inside the submission file before submitting the `result.zip` file. ### **Evaluation Method** The judging system uses a combination of two metrics, **Log Loss** and **RMSLE**, to evaluate the performance of your model. Since a value of zero in these metrics indicates zero error and a perfect prediction, the judging system calculates the score based on a **descending exponential function of your model's combined error**. The total score is **100 points**, where the error of each section is considered with specific weights (35% for correctly classifying the sales channel and 65% for predicting the revenue). The formula for calculating the final score is as follows: $$S = 100 \times e^{-1.3(0.35 \times LogLoss + 0.65 \times RMSLE)}$$ Where in this formula: %align_left_start% * **$LogLoss$**: The error of your model's logarithmic loss function for classifying the sales channel. * **$RMSLE$**: The root mean squared logarithmic error of your model for predicting revenue in USD. %align_end% The closer your model's error in both sections gets to zero, the closer the exponent gets to zero, and your final score will approach *100*.