%%{init: {'theme': 'base', 'themeVariables': {'primaryColor': '#efefef', 'primaryTextColor': '#2d2d2d', 'primaryBorderColor': '#cccccc', 'lineColor': '#999999', 'fontFamily': 'Inter, sans-serif', 'fontSize': '18px', 'edgeLabelBackground': '#ffffff'}, 'flowchart': {'rankDir': 'LR', 'curve': 'linear'}}}%%
flowchart LR
D["📊 Data"]
classDef accent fill:#e8925a,color:#fff,stroke:none
classDef muted fill:#efefef,color:#2d2d2d,stroke:#cccccc
class D muted
Week 5 — Feature Engineering
IAT 461 / 882 · Data Science for Human-Centered Systems · Summer 2026 · Alireza Karduni
Min-Max Scaling — code
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
from sklearn.preprocessing import MinMaxScaler
df = pd.read_csv("data/youtube_videos.csv")
scaler = MinMaxScaler()
df["subscribers_scaled"] = scaler.fit_transform(
df[["subscribers"]]
)
fig, axes = plt.subplots(2, 1, figsize=(4, 5))
sns.histplot(df["subscribers"],
ax=axes[0],
color="#e8925a", bins=30)
axes[0].set_title("Before")
axes[0].set_xlabel("subscribers")
sns.histplot(df["subscribers_scaled"],
ax=axes[1],
color="#6a9fb5", bins=30)
axes[1].set_title("After: Min-Max Scaled")
axes[1].set_xlabel("subscribers_scaled")
sns.despine()
plt.tight_layout()
plt.show()
Standardization — code
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
from sklearn.preprocessing import StandardScaler
df = pd.read_csv("data/youtube_videos.csv")
scaler = StandardScaler()
df["views_scaled"] = scaler.fit_transform(df[["views"]])
fig, axes = plt.subplots(2, 1, figsize=(4, 5))
sns.histplot(df["views"],
ax=axes[0],
color="#e8925a", bins=30)
axes[0].set_title("Before: views")
axes[0].set_xlabel("views")
sns.histplot(df["views_scaled"],
ax=axes[1],
color="#6a9fb5", bins=30)
axes[1].set_title("After: Standardized")
axes[1].set_xlabel("z-score")
sns.despine()
plt.tight_layout()
plt.show()
Robust Scaling — code
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
from sklearn.preprocessing import RobustScaler
df = pd.read_csv("data/youtube_videos.csv")
scaler = RobustScaler()
df["views_robust"] = scaler.fit_transform(df[["views"]])
fig, axes = plt.subplots(2, 1, figsize=(4, 5))
sns.histplot(df["views"],
ax=axes[0],
color="#e8925a", bins=30)
axes[0].set_title("Before: views")
axes[0].set_xlabel("views")
sns.histplot(df["views_robust"],
ax=axes[1],
color="#6a9fb5", bins=30)
axes[1].set_title("After: Robust Scaled")
axes[1].set_xlabel("robust scaled")
sns.despine()
plt.tight_layout()
plt.show()
Log Transform — code
import pandas as pd
import numpy as np
import seaborn as sns
import matplotlib.pyplot as plt
df = pd.read_csv("data/youtube_videos.csv")
df["log_views"] = np.log1p(df["views"])
fig, axes = plt.subplots(2, 1, figsize=(4, 5))
sns.histplot(df["views"],
ax=axes[0],
color="#e8925a", bins=30)
axes[0].set_title("Before: views")
axes[0].set_xlabel("views")
sns.histplot(df["log_views"],
ax=axes[1],
color="#6a9fb5", bins=30)
axes[1].set_title("After: log(views + 1)")
axes[1].set_xlabel("log_views")
sns.despine()
plt.tight_layout()
plt.show()
Binning — code
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
df = pd.read_csv("data/youtube_videos.csv")
bins = [0, 300, 1800, float("inf")]
labels = ["short", "medium", "long"]
df["duration_bin"] = pd.cut(
df["duration_seconds"],
bins=bins,
labels=labels
)
order = ["short", "medium", "long"]
colors = ["#6a9fb5", "#e8925a", "#7a7a7a"]
fig, axes = plt.subplots(2, 1, figsize=(4, 5))
sns.histplot(df["duration_seconds"],
ax=axes[0],
color="#e8925a", bins=30)
axes[0].set_title("Before: duration_seconds")
axes[0].set_xlabel("seconds")
sns.countplot(data=df,
x="duration_bin",
order=order,
palette=colors,
ax=axes[1])
axes[1].set_title("After: duration_bin")
axes[1].set_xlabel("duration bin")
axes[1].set_ylabel("count")
sns.despine()
plt.tight_layout()
plt.show()
Ordinal Encoding — code
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
from sklearn.preprocessing import OrdinalEncoder
df = pd.read_csv("data/youtube_videos.csv")
enc = OrdinalEncoder(categories=[["Small", "Mid", "Large"]])
df["channel_tier_encoded"] = enc.fit_transform(
df[["channel_tier"]]
)
fig, axes = plt.subplots(2, 1, figsize=(4, 5))
order_str = ["Small", "Mid", "Large"]
sns.countplot(data=df,
x="channel_tier",
order=order_str,
color="#e8925a",
ax=axes[0])
axes[0].set_title("Before: channel_tier")
axes[0].set_xlabel("tier")
order_num = [0, 1, 2]
sns.countplot(data=df,
x="channel_tier_encoded",
color="#6a9fb5",
ax=axes[1])
axes[1].set_title("After: encoded (0, 1, 2)")
axes[1].set_xlabel("encoded tier")
sns.despine()
plt.tight_layout()
plt.show()
Date & Time Features — code
import pandas as pd
import numpy as np
import seaborn as sns
import matplotlib.pyplot as plt
df = pd.read_csv("data/youtube_videos.csv")
# simulate a full timestamp from existing day/hour columns
day_map = {
"Monday": 0, "Tuesday": 1, "Wednesday": 2,
"Thursday": 3, "Friday": 4, "Saturday": 5, "Sunday": 6
}
rng = np.random.default_rng(42)
base = pd.Timestamp("2024-01-01")
df["uploaded_at"] = [
base
+ pd.offsets.Week(weekday=day_map[d])
+ pd.Timedelta(hours=int(h))
+ pd.Timedelta(minutes=int(rng.integers(0, 60)))
for d, h in zip(df["upload_day"], df["upload_hour"])
]
# extract features from timestamp
df["is_weekend"] = (df["uploaded_at"].dt.dayofweek >= 5).astype(int)
df["is_primetime"] = (
(df["uploaded_at"].dt.hour >= 18) &
(df["uploaded_at"].dt.hour <= 22)
).astype(int)
fig, axes = plt.subplots(1, 2, figsize=(4, 3))
sns.barplot(data=df,
x="is_weekend",
y="views",
color="#e8925a",
ax=axes[0])
axes[0].set_title("Weekend vs views")
axes[0].set_xlabel("is_weekend")
sns.barplot(data=df,
x="is_primetime",
y="views",
color="#6a9fb5",
ax=axes[1])
axes[1].set_title("Primetime vs views")
axes[1].set_xlabel("is_primetime")
sns.despine()
plt.tight_layout()
plt.show()
ALL CAPS count — code
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
df = pd.read_csv("data/youtube_videos.csv")
df["caps_count"] = df["title"].apply(
lambda t: sum(1 for w in t.split() if w.isupper())
)
fig, axes = plt.subplots(2, 1, figsize=(4, 5))
sns.histplot(df["caps_count"],
bins=8,
color="#e8925a",
ax=axes[0])
axes[0].set_title("Distribution of CAPS word count")
axes[0].set_xlabel("caps_count")
sns.boxplot(data=df,
x="caps_count",
y="views",
color="#6a9fb5",
ax=axes[1])
axes[1].set_title("CAPS count vs views")
axes[1].set_xlabel("caps_count")
axes[1].set_ylabel("views")
sns.despine()
plt.tight_layout()
plt.show()
TF-IDF — code
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
from sklearn.feature_extraction.text import TfidfVectorizer
df = pd.read_csv("data/youtube_videos.csv")
vectorizer = TfidfVectorizer(max_features=10, stop_words="english")
tfidf_matrix = vectorizer.fit_transform(df["title"])
terms = vectorizer.get_feature_names_out()
mean_scores = tfidf_matrix.toarray().mean(axis=0)
tfidf_df = pd.DataFrame({
"term": terms,
"mean_tfidf": mean_scores
}).sort_values("mean_tfidf", ascending=True)
sns.barplot(data=tfidf_df,
x="mean_tfidf",
y="term",
color="#6a9fb5")
plt.title("Top TF-IDF terms in video titles")
plt.xlabel("mean TF-IDF score")
plt.ylabel("")
sns.despine()
plt.tight_layout()
plt.show()
Retention Rate — code
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
import numpy as np
df = pd.read_csv("data/youtube_videos.csv")
df["duration_minutes"] = df["duration_seconds"] / 60
df["retention_rate"] = (
df["watch_time_minutes"] /
(df["views"] * df["duration_minutes"])
).clip(0, 1)
fig, axes = plt.subplots(2, 1, figsize=(4, 5))
sns.histplot(df["retention_rate"],
bins=30,
color="#e8925a",
ax=axes[0])
axes[0].set_title("Distribution of retention rate")
axes[0].set_xlabel("retention_rate")
sns.scatterplot(data=df,
x="retention_rate",
y="views",
hue="channel_tier",
palette={"Small": "#6a9fb5",
"Mid": "#e8925a",
"Large": "#7a7a7a"},
alpha=0.6,
ax=axes[1])
axes[1].set_yscale("log")
axes[1].set_title("Retention rate vs views")
axes[1].set_xlabel("retention_rate")
axes[1].set_ylabel("views (log scale)")
sns.despine()
plt.tight_layout()
plt.show()
Interaction Features — code
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
df = pd.read_csv("data/youtube_videos.csv")
df["reach_potential"] = (
df["subscribers"] * df["click_through_rate"]
)
fig, axes = plt.subplots(2, 1, figsize=(4, 5))
sns.scatterplot(data=df,
x="subscribers",
y="views",
color="#e8925a",
alpha=0.5,
ax=axes[0])
axes[0].set_xscale("log")
axes[0].set_yscale("log")
axes[0].set_title("subscribers vs views")
sns.scatterplot(data=df,
x="reach_potential",
y="views",
color="#6a9fb5",
alpha=0.5,
ax=axes[1])
axes[1].set_xscale("log")
axes[1].set_yscale("log")
axes[1].set_title("reach_potential vs views")
sns.despine()
plt.tight_layout()
plt.show()