Sounds like a plan. First step: gather the raw metrics—kills, deaths, assists, objective time, damage dealt, and the opponent’s skill rating. Then we’ll compute simple ratios like K/D, A/D, and objective control percentage. With those features we can fit a logistic regression or a random forest to predict win probability. Once we have the model, we’ll look at feature importance to see which stats actually drive the outcome. Let’s pull the data and get started.

Sure thing, let’s pull the dataset first. Load the match logs, calculate the key ratios—K/D, A/D, objective control—and feed them into a logistic model. Once we see the coefficients, we can tell which stats matter most. After that, we’ll run cross‑validation to estimate the prediction accuracy. I’ll get the code ready while you set up the mic.

Here’s a quick, ready‑to‑run snippet that pulls the raw match logs, computes the ratios, and fits a simple logistic regression to predict win probability. Feel free to swap out the data paths or add more features later. ```python # Imports import pandas as pd import numpy as np from sklearn.model_selection import train_test_split, cross_val_score from sklearn.preprocessing import StandardScaler from sklearn.linear_model import LogisticRegression from sklearn.metrics import roc_auc_score, accuracy_score # 1. Load raw match logs # Expected columns: ['match_id', 'team', 'kills', 'deaths', 'assists', # 'objective_time', 'damage_dealt', 'opponent_skill', 'win'] df = pd.read_csv('match_logs.csv') # 2. Compute key ratios and additional features df['k_d_ratio'] = df['kills'] / (df['deaths'] + 1) df['a_d_ratio'] = df['assists'] / (df['deaths'] + 1) df['damage_per_sec'] = df['damage_dealt'] / (df['objective_time'] + 1) df['objective_control_pct'] = df['objective_time'] / (df['objective_time'].sum() + 1e-6) # 3. Select predictors and target features = ['k_d_ratio', 'a_d_ratio', 'damage_per_sec', 'objective_control_pct', 'opponent_skill'] X = df[features] y = df['win'] # 1 for win, 0 for loss # 4. Scale predictors scaler = StandardScaler() X_scaled = scaler.fit_transform(X) # 5. Train-test split X_train, X_test, y_train, y_test = train_test_split( X_scaled, y, test_size=0.2, random_state=42, stratify=y ) # 6. Fit logistic regression logreg = LogisticRegression(max_iter=2000, solver='lbfgs') logreg.fit(X_train, y_train) # 7. Evaluate y_pred_prob = logreg.predict_proba(X_test)[:, 1] y_pred = logreg.predict(X_test) print('Accuracy:', accuracy_score(y_test, y_pred)) print('ROC AUC:', roc_auc_score(y_test, y_pred_prob)) # 8. Cross‑validation for robustness cv_scores = cross_val_score(logreg, X_scaled, y, cv=5, scoring='roc_auc') print('5‑fold CV AUC:', cv_scores.mean(), '+/-', cv_scores.std()) ``` **What the script does** 1. Reads a CSV of match logs (update the path as needed). 2. Calculates the ratios you mentioned plus a few extra features like damage per second and objective‑control percentage. 3. Standardises the predictors so the logistic regression isn’t biased toward any single scale. 4. Splits the data, trains the model, and prints basic metrics (accuracy and ROC AUC). 5. Runs a 5‑fold cross‑validation to give you a sense of how stable the performance is. You can extend the feature set with more nuanced stats—e.g., kill‑streak length, time‑to‑first‑kill, or map‑specific performance—then retrain and see how the AUC improves. Once you’ve identified the top predictors, those are the “MVP” stats you can brag about. Happy modeling!