ahmedheikal/AppliedMachineLearningProject

1import streamlit as st2import pandas as pd3import matplotlib.pyplot as plt4import seaborn as sns5import joblib6from pathlib import Path7 8st.set_page_config(9    page_title="Credit Card Customer Segmentation",10    layout="wide"11)12 13BASE = Path(".")14 15# =========================16# LOAD FILES17# =========================18@st.cache_data19def load_data():20    files = {21        "metrics": "evaluation_metrics.csv",22        "insights": "business_insights.csv",23        "processed": "data_scaled_with_kmeans.csv",24        "segments": "final_customer_segments.csv",25        "dbscan": "dbscan_labels.csv",26        "gmm_probs": "gmm_probabilities.csv"27    }28 29    data = {}30 31    for key, file in files.items():32        try:33            data[key] = pd.read_csv(BASE / file)34        except:35            data[key] = None36 37    return data38 39 40@st.cache_resource41def load_models():42    models = {}43 44    try:45        models["kmeans"] = joblib.load(BASE / "kmeans_model.pkl")46    except:47        models["kmeans"] = None48 49    try:50        models["gmm"] = joblib.load(BASE / "gmm_model.pkl")51    except:52        models["gmm"] = None53 54    try:55        models["scaler"] = joblib.load(BASE / "scaler.pkl")56    except:57        models["scaler"] = None58 59    try:60        models["pca"] = joblib.load(BASE / "pca_model.pkl")61    except:62        models["pca"] = None63 64    return models65 66 67data = load_data()68models = load_models()69 70# =========================71# HEADER72# =========================73st.title("Credit Card Customer Segmentation Dashboard")74st.markdown(75    """76Applied Machine Learning Project using:77 78- KMeans Clustering79- DBSCAN80- Gaussian Mixture Model (GMM)81- Deep Clustering using DEC82"""83)84 85# =========================86# TOP METRICS87# =========================88col1, col2, col3, col4 = st.columns(4)89 90with col1:91    st.metric("Models Used", "4")92 93with col2:94    if data["processed"] is not None:95        st.metric("Customers", len(data["processed"]))96    else:97        st.metric("Customers", "N/A")98 99with col3:100    st.metric("Best Deep Model", "DEC")101 102with col4:103    if data["metrics"] is not None:104        best_score = data["metrics"]["Silhouette"].max()105        st.metric("Best Silhouette", f"{best_score:.3f}")106    else:107        st.metric("Best Silhouette", "N/A")108 109st.divider()110 111# =========================112# PROJECT OVERVIEW113# =========================114left, right = st.columns([2, 1])115 116with left:117    st.subheader("Project Methodology")118 119    methodology = pd.DataFrame({120        "Stage": [121            "Preprocessing",122            "Feature Engineering",123            "Dimensionality Reduction",124            "Traditional Clustering",125            "Probabilistic Clustering",126            "Deep Clustering"127        ],128        "Description": [129            "Missing values, scaling, log transformation",130            "Behavioral financial feature creation",131            "PCA compression for efficient clustering",132            "KMeans and DBSCAN clustering",133            "Gaussian Mixture customer segmentation",134            "Autoencoder + DEC latent clustering"135        ]136    })137 138    st.dataframe(methodology, use_container_width=True)139 140with right:141    st.subheader("Dataset Overview")142 143    if data["processed"] is not None:144        st.write("Dataset Shape:")145        st.write(data["processed"].shape)146 147        st.write("Preview:")148        st.dataframe(data["processed"].head(), use_container_width=True)149    else:150        st.warning("Dataset file not found")151 152st.divider()153 154# =========================155# MODEL COMPARISON156# =========================157st.subheader("Model Performance Comparison")158 159if data["metrics"] is not None:160    st.dataframe(data["metrics"], use_container_width=True)161 162    fig = plt.figure(figsize=(10, 5))163    sns.barplot(164        data=data["metrics"],165        x="Model",166        y="Silhouette"167    )168    plt.xticks(rotation=30)169    plt.title("Silhouette Score Comparison")170    plt.tight_layout()171    st.pyplot(fig)172 173else:174    st.warning("evaluation_metrics.csv not found")175 176st.divider()177 178# =========================179# DBSCAN ANALYSIS180# =========================181col1, col2 = st.columns(2)182 183with col1:184    st.subheader("DBSCAN Analysis")185 186    if data["dbscan"] is not None:187        total = len(data["dbscan"])188        noise = (data["dbscan"]["DBSCAN_Cluster"] == -1).sum()189        clusters = data["dbscan"]["DBSCAN_Cluster"].nunique() - 1190 191        st.metric("Detected Clusters", clusters)192        st.metric("Noise Points", noise)193        st.metric("Noise Percentage", f"{(noise / total) * 100:.2f}%")194 195        st.info(196            "DBSCAN was useful for anomaly detection, but struggled due to overlapping density structures in customer behavior."197        )198    else:199        st.warning("dbscan_labels.csv not found")200 201with col2:202    st.subheader("GMM Confidence")203 204    if data["gmm_probs"] is not None:205        fig = plt.figure(figsize=(8, 4))206 207        numeric_probs = data["gmm_probs"].select_dtypes(include=["number"])208 209        sns.histplot(210            numeric_probs.max(axis=1),211            bins=30212        )213 214        plt.title("GMM Assignment Confidence")215        plt.xlabel("Maximum Cluster Probability")216        plt.tight_layout()217 218        st.pyplot(fig)219 220    else:221        st.info("gmm_probabilities.csv not found")222 223st.divider()224 225# =========================226# BUSINESS INSIGHTS227# =========================228st.subheader("Business Customer Segmentation Insights")229 230if data["insights"] is not None:231    st.dataframe(data["insights"], use_container_width=True)232else:233    st.warning("business_insights.csv not found")234 235st.divider()236 237# =========================238# FINAL DEC SEGMENTS239# =========================240st.subheader("Final Deep Clustering Segmentation (DEC)")241 242if data["segments"] is not None:243    st.dataframe(data["segments"].head(20), use_container_width=True)244 245    if "Customer_Type" in data["segments"].columns:246        counts = data["segments"]["Customer_Type"].value_counts()247 248        fig = plt.figure(figsize=(8, 5))249        counts.plot(kind="bar")250        plt.title("Customer Segment Distribution")251        plt.xticks(rotation=30)252        plt.tight_layout()253        st.pyplot(fig)254 255else:256    st.warning("final_customer_segments.csv not found")257 258st.divider()259 260# =========================261# FINAL RECOMMENDATION262# =========================263st.subheader("Final Recommendation")264 265recommendation = pd.DataFrame({266    "Category": [267        "Best Traditional Model",268        "Best Density Model",269        "Best Probabilistic Model",270        "Best Deep Learning Model",271        "Recommended Final Approach"272    ],273    "Model": [274        "KMeans",275        "DBSCAN",276        "GMM",277        "DEC",278        "DEC"279    ]280})281 282st.dataframe(recommendation, use_container_width=True)283 284st.success(285    """286Deep Clustering using DEC achieved the strongest customer separation performance.287This makes it the most effective segmentation approach for this project.288"""289)