Data Visualisation & Predictive Analytics Dashboard

01. The Real Problem

Analyzing global socio-economic, environmental, and PESTLE (Political, Economic, Social, Technological, Legal, Environmental) event streams typically requires juggling separate tools for data munging, static reporting, and statistical modeling. When decision-makers explore complex indicators like intensity, relevance, and likelihood, static tables fail to convey correlations or allow speculative 'what-if' forecasting.

The goal of this project was to build a unified exploratory analytics workspace in Python that combines high-performance multi-variable filtering, dynamic charting, and real-time regression forecasting in a single accessible web interface.

02. Data Processing & Feature Pipeline

In tt.py, raw data from jsondata.json is ingested and prepared with robust cleaning steps:

# Handle missing values and adjust data types
df['end_year'] = pd.to_numeric(df['end_year'], errors='coerce').fillna(0).astype(int)
df['intensity'] = pd.to_numeric(df['intensity'], errors='coerce').fillna(0)
df['relevance'] = pd.to_numeric(df['relevance'], errors='coerce').fillna(0)
df['likelihood'] = pd.to_numeric(df['likelihood'], errors='coerce').fillna(0)
for col in ['topic', 'sector', 'region', 'pestle', 'source', 'country']:
    df[col] = df[col].fillna('Unknown')

# Encode categorical variables for the ML model
label_encoders = {}
for column in ['topic', 'sector', 'region', 'pestle', 'source', 'country']:
    le = LabelEncoder()
    df[column + '_encoded'] = le.fit_transform(df[column].astype(str))
    label_encoders[column] = le

Seven sidebar dropdowns allow visitors to slice the dataset dynamically across End Year, Topic, Sector, Region, PESTLE, Source, and Country, with instantaneous chart rerendering.

03. 6 Verified Interactive Visualizations

The dashboard implements 6 distinct Plotly visualizations designed for deep analytical exploration:

04. Random Forest Predictive Engine

Beyond retrospective charting, the application embeds a trained regression model to forecast the likelihood score of unseen events:

• Model Architecture: RandomForestRegressor(n_estimators=100, random_state=42) trained on an 80/20 train/test split.
• Evaluation Metric: Mean Absolute Error (MAE) evaluated on the held-out test split and displayed transparently to the user.
• Interactive Forecasting Form: Visitors adjust numeric sliders (Intensity, Relevance, Target Year) and categorical dropdowns (Sector, Region, PESTLE, Topic, Source, Country). The inputs are mapped through the fitted label_encoders and submitted to model.predict([encoded_values]), outputting the forecasted likelihood instantly.

05. Engineering Decisions & Optimization Scope