Unlocking PUBG Victory: Data‑Driven Insights on Drop Zones, Final Circles, Weapons, and Kill Strategies

1. Dataset Description

The analysis uses two Kaggle datasets from PUBG: one containing match‑level statistics and another with player‑kill records, each about 10 GB. The death data records every player death in roughly 720 k matches, while the aggregate data summarizes per‑match metadata and player statistics such as kills, damage, and distance traveled.

2. Where to land to avoid early death?

Heatmaps of player deaths within the first four minutes show dangerous zones (red areas) on both Erangel and Miramar maps. Safer, resource‑rich zones include coastal and bridge‑head locations. The following Python code loads the CSV files, filters deaths occurring under 240 seconds, scales coordinates, and generates heatmaps over the map images.

import pandas as pd
from scipy.ndimage.filters import gaussian_filter
import matplotlib.cm as cm
from matplotlib.colors import Normalize
from scipy.misc.pilutil import imread
import numpy as np
import matplotlib.pyplot as plt

# Load death data
death_0 = pd.read_csv('F:/pubg-match-deaths/deaths/kill_match_stats_final_0.csv')
death_1 = pd.read_csv('F:/pubg-match-deaths/deaths/kill_match_stats_final_1.csv', nrows=5000000)
death = death_0.merge(death_1, how='outer')

# Filter early deaths (<240 s) for each map
in_240_seconds_erg = death.loc[(death['map'] == 'ERANGEL') & (death['time'] < 240)].dropna()
in_240_seconds_mrm = death.loc[(death['map'] == 'MIRAMAR') & (death['time'] < 240)].dropna()

# Scale coordinates
data_erg = in_240_seconds_erg[['victim_position_x','victim_position_y']].values * 4096 / 800000
data_mrm = in_240_seconds_mrm[['victim_position_x','victim_position_y']].values * 1000 / 800000

def heatmap(x, y, s, bins=100):
    heatmap, xedges, yedges = np.histogram2d(x, y, bins=bins)
    heatmap = gaussian_filter(heatmap, sigma=s)
    extent = [xedges[0], xedges[-1], yedges[0], yedges[-1]]
    return heatmap.T, extent

# Erangel heatmap
bg = imread('F:/pubg-match-deaths/erangel.jpg')
hmap, extent = heatmap(data_erg[:,0], data_erg[:,1], 4.5)
alphas = np.clip(Normalize(0, hmap.max(), clip=True)(hmap) * 4.5, 0.0, 1.)
colors = cm.Reds(Normalize(0, hmap.max(), clip=True)(hmap))
colors[..., -1] = alphas
fig, ax = plt.subplots(figsize=(24,24))
ax.set_xlim(0,4096); ax.set_ylim(0,4096)
ax.imshow(bg); ax.imshow(colors, extent=extent, origin='lower', cmap=cm.Reds, alpha=0.9)
plt.gca().invert_yaxis()
plt.savefig('out1.png', dpi=100)

# Miramar heatmap (similar steps)
bg = imread('F:/pubg-match-deaths/miramar.jpg')
hmap, extent = heatmap(data_mrm[:,0], data_mrm[:,1], 4.5)
alphas = np.clip(Normalize(0, hmap.max(), clip=True)(hmap) * 4.5, 0.0, 1.)
colors = cm.Reds(Normalize(0, hmap.max(), clip=True)(hmap))
colors[..., -1] = alphas
fig, ax = plt.subplots(figsize=(24,24))
ax.set_xlim(0,1000); ax.set_ylim(0,1000)
ax.imshow(bg); ax.imshow(colors, extent=extent, origin='lower', cmap=cm.Reds, alpha=0.9)
plt.gca().invert_yaxis()
plt.savefig('out2.png', dpi=100)

3. Best final‑circle locations

Heatmaps of the positions of the final‑placement (rank 1) players show that high‑probability win zones are broadly distributed but concentrate around map centers and former drop‑hotspots. This suggests that staying near the center after the first circle collapse improves survival odds.

last_seconds_erg = death.loc[(death['map'] == 'ERANGEL') & (death['killer_placement'] == 1)].dropna()
last_seconds_mrm = death.loc[(death['map'] == 'MIRAMAR') & (death['killer_placement'] == 1)].dropna()

data_erg = last_seconds_erg[['victim_position_x','victim_position_y']].values * 4096 / 800000
data_mrm = last_seconds_mrm[['victim_position_x','victim_position_y']].values * 1000 / 800000

# Erangel final‑circle heatmap (same function as before)
bg = imread('F:/pubg-match-deaths/erangel.jpg')
hmap, extent = heatmap(data_erg[:,0], data_erg[:,1], 4.5)
# ... (plotting steps identical to previous section) ...
plt.savefig('out3.png', dpi=100)

# Miramar final‑circle heatmap
bg = imread('F:/pubg-match-deaths/miramar.jpg')
hmap, extent = heatmap(data_mrm[:,0], data_mrm[:,1], 4.5)
# ... (plotting steps) ...
plt.savefig('out4.png', dpi=100)

4. Which weapons boost win chances?

Analysis of the weapon distribution among winning players shows the automatic rifle M416 as the most frequently used, followed by the sniper rifle 98‑k. The data suggests that players who win tend to favor versatile close‑to‑mid‑range weapons over pure snipers.

5. Distance vs. win probability

Kill‑distance histograms reveal that most kills by winners occur within 500 m, indicating a preference for close‑range engagements. Long‑range sniper kills (>5 km) are rare and typically involve high‑magnification scopes.

last_seconds_erg = death.loc[death['killer_placement'] == 1].dropna()

distance = np.sqrt(((last_seconds_erg['killer_position_x'] - last_seconds_erg['victim_position_x']) / 100) ** 2 +
                   ((last_seconds_erg['killer_position_y'] - last_seconds_erg['victim_position_y']) / 100) ** 2)

distance = distance.apply(lambda x: int(x))
labels = [0,10,30,50,100,200,500,800,1000,1500,2000,3000,5000,10000,20000]

dist_cut = pd.cut(distance, bins=labels)
dist_cut.value_counts().plot.bar(figsize=(10,8))
plt.savefig('out6.png', dpi=100)

6. Kill count vs. win probability

Examining the kill distribution of match winners shows that most champions secure two or fewer kills, contradicting the belief that high kill counts guarantee victory. This highlights the importance of strategic positioning and survival over pure aggression.

match_stats = pd.read_csv('F:/pubg-match-deaths/aggregate/agg_match_stats_0.csv')
winner = match_stats.loc[match_stats['team_placement'] == 1].dropna()
labels = [0,2,5,8,11,15,20,30,40,50]
winner['kill'] = pd.cut(winner['player_kills'], bins=labels)
winner['assist'] = pd.cut(winner['player_assists'], bins=labels)

winner['kill'].value_counts().plot.bar(figsize=(10,10))
plt.savefig('out7.png', dpi=100)