Quick start

  1. Open the Dashboard. Use the sidebar to access the prediction form.
  2. Fill in game features. Enter pre-launch attributes across the 10 sections (Pricing, Release, Genre, …).
  3. Submit. Click Predict Success to run the ensemble model.
  4. Review. Inspect the predicted owner tier, confidence, class probabilities, and recommendations.

Input reference

All non-composite features, grouped by section:

💰 Pricing

Field Label Type Range
is_free Free to Play? Yes / No 0 / 1
price Price (0 - 200 USD) Number 0–200
initialprice Initial Price (0 - 200 USD) Number 0–200

🗓️ Release

Field Label Type Range
release_date Release Date Date picker YYYY-MM-DD

🎮 Genre

Field Label Type Range
Action Action Yes / No 0 / 1
Adventure Adventure Yes / No 0 / 1
RPG RPG Yes / No 0 / 1
Strategy Strategy Yes / No 0 / 1
Simulation Simulation Yes / No 0 / 1
Sports Sports Yes / No 0 / 1
Racing Racing Yes / No 0 / 1

🖥️ Platform

Field Label Type Range
platform_windows Windows Yes / No 0 / 1
platform_mac Mac Yes / No 0 / 1
platform_linux Linux Yes / No 0 / 1
platform_count Total Platforms (auto) Number 1–3

🌍 Languages

Field Label Type Range
selected_languages Supported Languages Language checklist 20 languages
supported_languages_count Text Languages (auto-count) Number 0–50
full_audio_languages_count Full Audio Languages (0 - 20) Number 0–20

🏪 Store Page

Field Label Type Range
has_website Has Official Website? Yes / No 0 / 1
has_support_email Has Support Email? Yes / No 0 / 1
screenshot_count Number of Screenshots (0 - 20) Number 0–20
about_length Description Length (0 - 5000 chars) Number 0–5000
has_detailed_desc Detailed Description (auto) Yes / No 0 / 1

🏆 Categories

Field Label Type Range
has_achievements Steam Achievements? Yes / No 0 / 1
has_cloud_save Steam Cloud Save? Yes / No 0 / 1
has_controller_support Controller Support? Yes / No 0 / 1
has_vr_support VR Support? Yes / No 0 / 1
has_in_app_purchases In-App Purchases? Yes / No 0 / 1
has_family_sharing Family Sharing? Yes / No 0 / 1
category_count Total Steam Categories (auto) Number 0–15
achievement_count Number of Achievements (0 - 500) Number 0–500

🎯 Audience

Field Label Type Range
is_multiplayer Multiplayer Game? Yes / No 0 / 1
is_mature_content Mature Content? Yes / No 0 / 1
required_age Required Age (auto, derived from mature content) Number 0–17

🏷️ Tags

Field Label Type Range
selected_tags Steam Tags (select all that apply) Tag checklist 50 tags
tag_count Tag Count (auto) Number 0–20

📦 Packaging

Field Label Type Range
package_count Package Count (1 - 10) Number 1–10
sku_count SKU Count (1 - 20) Number 1–20

Understanding results

Predicted owner tier

  • Class 0 (≤10K) — Common Indie
  • Class 1 (35K) — Niche
  • Class 2 (75K) — Growing
  • Class 3 (150K) — Established
  • Class 4 (350K) — Popular
  • Class 5 (≥750K) — Breakout Hit

Confidence score

  • 80–100%: very confident
  • 60–79%: moderately confident
  • 40–59%: low confidence (borderline)
  • Below 40%: uncertain (mixed/weak features)

Class probabilities

Predicted probability for each of the 6 classes. The class with the highest probability is the final prediction.

Recommendations

  • Strengths — features currently working in your favor (high positive impact)
  • Improvements — actionable suggestions to lift the predicted tier (driven by SHAP analysis)

Best practices

  • Be accurate. The model is trained on real Steam data; inflated numbers mislead predictions.
  • Pre-launch only. No reviews or playtime — fill only what you can decide before release.
  • Use recommendations. Prioritise high-impact suggestions surfaced by SHAP.
  • Test scenarios. Try with/without publisher, varying pricing, etc., to see what moves the needle.
  • Don't over-optimise. Correlations ≠ causation. Game quality is the ultimate factor.

FAQ

Can I predict for unreleased games?
Yes — the model uses only pre-launch features (pricing, store page, platform support, tags, languages, etc.).
How do I set the release date?
Use the date picker to set your planned launch date. SAGE automatically derives release timing signals from it — such as the release month, quarter, day of week, and whether it falls in a summer or holiday window — which the model uses as predictive features.
What if some features are unknown?
Use the defaults shown in the form. Leave toggles unchecked (0) for features you don't plan to ship. Leave language/tag checklists empty if you haven't decided yet.
Why is my confidence low?
Your features fall in regions where multiple classes overlap. Adjust key drivers (store page quality, pricing, platform reach, language coverage) and try again.
How accurate is the model?
The ensemble achieves 71.20% accuracy and 0.6170 weighted F1 on the test set. See Model Metrics for details.
Does it work for non-Steam platforms?
No. Trained exclusively on Steam data; predictions for Epic, GOG, or consoles aren't reliable.