🧠 AI with Python – 📊 Building a Simple ML Dashboard (Streamlit Metrics)

Training and deploying a machine learning model is only part of the journey. Once a model is in production, teams need visibility into how it is performing over time.

Questions such as:

• Is accuracy declining?
• Are predictions increasing?
• Is model confidence changing?
• Do we need retraining?

cannot be answered by the model alone.

This is where ML dashboards become valuable.

In this project, we build a simple machine learning monitoring dashboard using Streamlit, one of the most popular frameworks for rapidly creating data applications in Python.

---

Why ML Dashboards Matter

Imagine a model deployed in production six months ago. The model continues generating predictions every day. Without monitoring, the team has no easy way to know:

• whether performance is stable
• whether prediction volume is changing
• whether confidence scores are decreasing
• whether concept drift is occurring

A dashboard provides a centralised view of model health.

---

What Is an ML Dashboard?

An ML dashboard is a visual interface that displays important model metrics in real time or near real time.

Typical information includes:

• Accuracy
• Precision
• Recall
• F1 Score
• Prediction Volume
• Confidence Scores
• Drift Indicators

These metrics help teams understand how models behave after deployment.

---

Why Streamlit?

Streamlit makes it possible to create dashboards using pure Python.

Instead of building:

• frontend applications
• APIs
• JavaScript components

developers can create interactive dashboards with only a few lines of code.

Example:

st.title("ML Model Monitoring Dashboard")

A complete web application is generated automatically.

---

Creating Monitoring Data

For this example, we simulate model monitoring metrics.

monitoring_data = pd.DataFrame(...)

The dataset contains:

• daily accuracy
• precision
• recall
• F1 score
• prediction volume
• confidence scores

These values mimic production monitoring information.

---

Displaying Key Metrics

The first section of the dashboard highlights current model performance.

col1.metric(
    "Accuracy",
    latest["accuracy"]
)

Streamlit’s metric component creates compact KPI cards.

Typical dashboards display:

• current accuracy
• current precision
• current recall
• current F1 score

at the top of the page.

---

Visualising Accuracy Trends

Metrics become more meaningful when viewed over time.

ax.plot(
    monitoring_data["date"],
    monitoring_data["accuracy"]
)

Trend charts help answer questions such as:

• Is accuracy stable?
• Is performance degrading?
• Did accuracy suddenly change?

Visualisation often reveals issues that raw numbers cannot.

---

Monitoring Prediction Volume

Prediction volume is another useful signal.

ax.plot(
    monitoring_data["prediction_volume"]
)

Unexpected changes in volume may indicate:

• traffic spikes
• system failures
• upstream pipeline issues

Monitoring usage patterns helps teams identify operational problems.

---

Adding Alerts

Production dashboards often include alerting logic.

if latest["accuracy"] < 0.90:
    st.warning(...)

Alerts allow teams to react quickly when performance drops below acceptable thresholds.

Examples include:

• declining accuracy
• increasing latency
• confidence degradation
• drift detection events

---

Displaying Raw Data

Dashboards often include underlying monitoring records.

st.dataframe(
    monitoring_data
)

This provides transparency and supports deeper investigation. Users can inspect the actual values behind visualisations.

---

What Real ML Dashboards Monitor

Production systems often track much more than accuracy. Common metrics include:

Model Metrics

• accuracy
• precision
• recall
• F1 score
• ROC-AUC

Operational Metrics

• prediction volume
• latency
• throughput
• error rates

Data Metrics

• feature drift
• concept drift
• missing value rates
• distribution changes

Modern MLOps platforms combine all of these into a unified dashboard.

---

Where ML Dashboards Are Used

Monitoring dashboards are common in:

• fraud detection systems
• recommendation engines
• healthcare ML platforms
• financial risk systems
• customer analytics products

Any production ML system benefits from visibility into performance and usage.

---

Advantages of Streamlit Dashboards

Streamlit offers several benefits:

• rapid development
• minimal frontend knowledge required
• interactive visualisations
• Python-native workflow
• easy deployment

This makes it an excellent choice for ML engineers and data scientists.

---

Key Takeaways

1. ML dashboards provide visibility into production model health.
2. Streamlit allows dashboards to be built using only Python.
3. Metrics such as accuracy, precision, recall, and F1-score are commonly monitored.
4. Trend visualisations help identify degradation and anomalies.
5. Dashboards are a foundational component of modern MLOps practices.

---

Conclusion

Deploying a machine learning model without monitoring is like driving a car without a dashboard. Teams need visibility into performance, usage patterns, and potential issues to keep models reliable over time. Streamlit provides a fast and practical way to build monitoring dashboards that help organisations understand and maintain production ML systems.

---

# 📦 Import Required Libraries
import streamlit as st
import pandas as pd
import matplotlib.pyplot as plt


# =========================================================
# 📝 Create Sample Monitoring Data
# =========================================================

monitoring_data = pd.DataFrame({
    "date": pd.date_range(
        start="2025-01-01",
        periods=10,
        freq="D"
    ),

    "accuracy": [
        0.96, 0.95, 0.95, 0.94, 0.93,
        0.92, 0.91, 0.90, 0.89, 0.88
    ],

    "precision": [
        0.95, 0.94, 0.94, 0.93, 0.92,
        0.91, 0.90, 0.89, 0.88, 0.87
    ],

    "recall": [
        0.94, 0.94, 0.93, 0.92, 0.91,
        0.90, 0.89, 0.88, 0.87, 0.86
    ],

    "f1_score": [
        0.945, 0.94, 0.935, 0.925, 0.915,
        0.905, 0.895, 0.885, 0.875, 0.865
    ],

    "prediction_volume": [
        500, 520, 510, 540, 560,
        580, 600, 610, 630, 650
    ],

    "avg_confidence": [
        0.94, 0.94, 0.93, 0.92, 0.91,
        0.90, 0.89, 0.88, 0.87, 0.86
    ]
})


# =========================================================
# 🖥️ Streamlit Dashboard Title
# =========================================================

st.title("ML Model Monitoring Dashboard")

st.write(
    "Simple dashboard to track model performance metrics over time."
)


# =========================================================
# 📊 Display Latest Metrics
# =========================================================

latest = monitoring_data.iloc[-1]

col1, col2, col3, col4 = st.columns(4)

col1.metric(
    "Accuracy",
    round(latest["accuracy"], 3)
)

col2.metric(
    "Precision",
    round(latest["precision"], 3)
)

col3.metric(
    "Recall",
    round(latest["recall"], 3)
)

col4.metric(
    "F1 Score",
    round(latest["f1_score"], 3)
)


# =========================================================
# 📈 Accuracy Trend
# =========================================================

st.subheader("Accuracy Trend")

fig, ax = plt.subplots(figsize=(8, 4))

ax.plot(
    monitoring_data["date"],
    monitoring_data["accuracy"],
    marker="o"
)

ax.set_title("Model Accuracy Over Time")
ax.set_xlabel("Date")
ax.set_ylabel("Accuracy")
ax.grid(True)

st.pyplot(fig)


# =========================================================
# 📈 Classification Metrics Trend
# =========================================================

st.subheader("Classification Metrics Trend")

fig, ax = plt.subplots(figsize=(8, 4))

ax.plot(
    monitoring_data["date"],
    monitoring_data["precision"],
    marker="o",
    label="Precision"
)

ax.plot(
    monitoring_data["date"],
    monitoring_data["recall"],
    marker="o",
    label="Recall"
)

ax.plot(
    monitoring_data["date"],
    monitoring_data["f1_score"],
    marker="o",
    label="F1 Score"
)

ax.set_title("Precision, Recall, and F1-score Over Time")
ax.set_xlabel("Date")
ax.set_ylabel("Score")
ax.legend()
ax.grid(True)

st.pyplot(fig)


# =========================================================
# 📈 Prediction Volume Trend
# =========================================================

st.subheader("Prediction Volume Trend")

fig, ax = plt.subplots(figsize=(8, 4))

ax.plot(
    monitoring_data["date"],
    monitoring_data["prediction_volume"],
    marker="o"
)

ax.set_title("Prediction Volume Over Time")
ax.set_xlabel("Date")
ax.set_ylabel("Prediction Count")
ax.grid(True)

st.pyplot(fig)


# =========================================================
# 📈 Average Confidence Trend
# =========================================================

st.subheader("Average Confidence Trend")

fig, ax = plt.subplots(figsize=(8, 4))

ax.plot(
    monitoring_data["date"],
    monitoring_data["avg_confidence"],
    marker="o"
)

ax.set_title("Average Prediction Confidence Over Time")
ax.set_xlabel("Date")
ax.set_ylabel("Average Confidence")
ax.grid(True)

st.pyplot(fig)


# =========================================================
# 🚨 Simple Monitoring Alerts
# =========================================================

st.subheader("Monitoring Alerts")

if latest["accuracy"] < 0.90:
    st.warning("⚠️ Accuracy has dropped below threshold.")
else:
    st.success("✅ Accuracy is within healthy range.")

if latest["avg_confidence"] < 0.87:
    st.warning("⚠️ Average confidence is decreasing.")
else:
    st.success("✅ Average confidence is within healthy range.")


# =========================================================
# 📋 Display Raw Monitoring Data
# =========================================================

st.subheader("Monitoring Data")

st.dataframe(monitoring_data)


# =========================================================
# ▶️ Run Command
# =========================================================

st.info(
    "Run this dashboard using: streamlit run ml_dashboard.py"
)