Day 24: Data Odyssey – What is Time Series Analysis?

Welcome to Day 24: Data Odyssey, our 365-day journey to master data science and artificial intelligence (AI), launched on Shivaratri, February 26, 2025! Yesterday, in Day 23: Data Odyssey – How Do We Deploy ML Models?, we brought Priya’s Random Forest models to life. Saved as .pkl files, her regression (MAE ₹4) predicted ₹642 for Thursday’s 9 AM Samosa sales, and her classifier (95% cross-val) flagged “Busy”—all via a live script using her 7-row dataset. Today, we shift to time: What is time series analysis, and how can Priya uncover trends in her growing sales data?

The Rhythm of Time

Time series analysis studies data ordered by time—Priya’s sales at 7 AM, 8 AM, 9 AM across days. Unlike her Random Forest (Day 22), which treated rows independently, time series respects sequence: Monday’s ₹600 at 9 AM leads to Tuesday’s ₹650. It’s an “analyze” tool in our workflow (Day 1), spotting patterns—daily peaks, weekly cycles—for better forecasts.

Think of it as Priya tracking her café’s pulse. Sales aren’t random; they flow—8-9 AM surges, rainy days dip. Day 24: Data Odyssey taps this rhythm.

Why Time Series Matters

Priya’s deployed model predicts ₹642 for 9 AM—but why ₹642? Time series answers:

• Trends: Are sales rising weekly?
• Seasonality: Does 9 AM peak daily?
• Forecasts: What’s next week’s 9 AM?

Her 7 rows hint at 8-9 AM rushes (Day 6’s EDA), but time series scales this to weeks, months—stock smarter, not just today. Day 24: Data Odyssey reveals this.

Priya’s Data as Time Series

Her 7 rows (Day 23), ordered:

   Date       Hour_Num  Item_Code  Sales
0  Monday     7          0        200
1  Monday     8          0        500
2  Monday     9          1        600
3  Tuesday    7          0        150
4  Tuesday    8          0        550
5  Tuesday    9          1        650
6  Wednesday  9          1        640

• Time: Date + Hour_Num.
• Value: Sales.
• Gaps: Missing 10-11 AM—assume 0 or fill later.

Goal: Analyze trends—9 AM’s rise? Day 24: Data Odyssey starts here.

Time Series Basics

Key parts:

1. Trend: Long-term rise/fall—sales growing?
2. Seasonality: Daily cycles—8-9 AM peaks?
3. Noise: Random dips—Tuesday’s 7 AM ₹150.

Pandas handles this—time as index. Day 24: Data Odyssey structures it.

Setting Up

Convert to time series:

import pandas as pd

# Data with datetime
data = pd.DataFrame({
    "Date": ["2025-03-03", "2025-03-03", "2025-03-03", "2025-03-04", "2025-03-04", "2025-03-04", "2025-03-05"],
    "Hour_Num": [7, 8, 9, 7, 8, 9, 9],
    "Item_Code": [0, 0, 1, 0, 0, 1, 1],
    "Sales": [200, 500, 600, 150, 550, 650, 640]
})
data["Datetime"] = pd.to_datetime(data["Date"]) + pd.to_timedelta(data["Hour_Num"], unit="h")
data.set_index("Datetime", inplace=True)
print(data[["Sales"]])

Output:

                     Sales
2025-03-03 07:00:00    200
2025-03-03 08:00:00    500
2025-03-03 09:00:00    600
2025-03-04 07:00:00    150
2025-03-04 08:00:00    550
2025-03-04 09:00:00    650
2025-03-05 09:00:00    640

Time-indexed—ready! Day 24: Data Odyssey times this.

Visualizing Trends

Plot it:

import matplotlib.pyplot as plt
data["Sales"].plot(figsize=(10, 6), marker="o", color="teal")
plt.title("Priya’s Sales Over Time")
plt.xlabel("Date and Hour")
plt.ylabel("Sales (₹)")
plt.grid(True)
plt.show()

• Pattern: 7 AM low (₹150-200), 8-9 AM high (₹500-650).
• Trend: Slight 9 AM rise—₹600 to ₹650.
• Gaps: Missing hours—sparse.

8-9 AM jumps daily—seasonality? Day 24: Data Odyssey sees this.

Daily Aggregation

Sum by day:

daily = data["Sales"].resample("D").sum()
print(daily)
daily.plot(marker="o", color="teal")
plt.title("Daily Total Sales")
plt.xlabel("Date")
plt.ylabel("Sales (₹)")
plt.grid(True)
plt.show()

Output:

2025-03-03    1300
2025-03-04    1350
2025-03-05     640

• Monday: ₹1300.
• Tuesday: ₹1350—up!
• Wednesday: ₹640—9 AM only.

Trend up, then dip—data’s short. Day 24: Data Odyssey sums this.

Hourly Patterns

Average by hour:

hourly_avg = data.groupby("Hour_Num")["Sales"].mean()
print(hourly_avg)
hourly_avg.plot(kind="bar", color="teal")
plt.title("Average Sales by Hour")
plt.xlabel("Hour")
plt.ylabel("Avg Sales (₹)")
plt.show()

Output:

Hour_Num
7    175.0
8    525.0
9    630.0

• 9 AM: ₹630 avg—rush king.
• 8 AM: ₹525—strong.
• 7 AM: ₹175—quiet.

Daily cycle—9 AM reigns. Day 24: Data Odyssey bars this.

Decomposition

Split trend, seasonality (needs more data, but try):

from statsmodels.tsa.seasonal import seasonal_decompose
result = seasonal_decompose(data["Sales"], model="additive", period=3)  # 3-hour cycle
result.plot()
plt.show()

• Trend: Slight rise—₹600 to ₹650.
• Seasonal: 7-8-9 AM pattern.
• Residual: Noise—Tuesday’s ₹150 dip.

7 rows limit clarity—35 rows (Day 12) sharpen it. Day 24: Data Odyssey breaks this.

Why Time Series?

• Context: ₹642 isn’t random—9 AM trend.
• Forecast: Next 9 AM—₹650+?
• Plan: Stock for rushes, not guesses.

Priya’s Random Forest (Day 23) lacks time—series adds it. Day 24: Data Odyssey times her.

Real-World Time Series

India’s power grid analyzes usage—peaks managed. Amazon tracks daily sales—stock aligns. Priya’s 7-row series is her café’s beat—small, growing. Day 24: Data Odyssey connects this.

Challenges

• Sparse: 7 points—gaps (10-11 AM) blur.
• Short: 3 days—no weekly cycle.
• Noise: ₹150 outlier—smooth it?

Day 12’s 35 rows fill gaps—Priya scales. Day 24: Data Odyssey flags this.

Why This Matters

Time series shows Priya’s 9 AM ₹630 avg—stock 40 samosas daily, not random ₹642. Without it, she misses cycles; with it, she plans—profit up. Scale it: time series predicts India’s monsoons—crops thrive. Day 24: Data Odyssey rhythms her.

Recap Summary

Yesterday, Day 23: Data Odyssey deployed Priya’s models—Random Forest predicted ₹642, “Busy” live. Today, Day 24: Data Odyssey introduced time series—9 AM ₹630 avg, daily cycles in 7 rows. It’s her time step.

What’s Next

Tomorrow, in Day 25: Data Odyssey – How Do We Forecast with Time Series?, we’ll predict: What’s Priya’s next 9 AM? We’ll use her series with simple forecasting, aiming past ₹642. Bring your curiosity, and I’ll see you there!