Churn Analysis with Matchine Learning

I built a churn predictor using Logistic Regression from the skit-learn library package, trained on 1,200 players.

Main Goals of the day:

  • Predict who will churn in the next 7 days (who will leave the gaming platform)
  • Use only features I can measure: sessions, days since last play, deposit, bonus
  • Output: a ranked list of “high-risk” players

Step by Step

đź“Ť Step 1: Aggregated player-level features:

  • total_sessions
  • last_session_days_ago
  • first_deposit
  • has_bonus
  • total_ggr

đź“Ť Step 2: Created label: is_high_risk = 1 if:

  • sessions < 3
  • last_session > 3 days ago
  • deposit < €50
  • no bonus claimed

đź“Ť Step 3: Trained LogisticRegression(class_weight='balanced')
📍 Step 4: AUC = 0.87 → strong model

Challenges / Insights

Teaching the machine how to analyze the players without using cycles or functions.

The purpose of the model isn’t for “predicting the future”, but for preventing the potential loss.

It found 216 players at risk.
Profile:

  • Played 2 times
  • Last played 4 days ago
  • Deposited €28
  • Never used a bonus

What do I do? Send them a “CASINO20” bonus to check if they return.

It’s common sense — automated.

Code Snippet Final

model = LogisticRegression(class_weight='balanced')
model.fit(X, y)
y_proba = model.predict_proba(X_test)[:, 1]
top_risk = df.sort_values('churn_probability', ascending=False).head(100)
top_risk.to_csv('reports/top_churn_risk_players.csv')

</pre>

Next Step

👉 Tableau report → Time to visualize the results.