MALANG TOURISM ROUTE OPTIMIZATION.
GENETIC ALGORITHM, NOT AN LLM.

Not every problem needs a slow, hallucination-prone LLM that burns cloud compute. Daily trip routing is a mathematical optimization problem — solved with a Genetic Algorithm that runs in <200 milliseconds entirely in your browser.

⚡ LAUNCH TRIP PLANNER (INTERACTIVE APP) WHY NOT AN LLM? 📄 READ THE PAPER (J-PTIIK)

100 Tourist Destinations

50 Indexed Restaurants

<200ms Execution Time

100% Client-Side

// PAPER SUMMARY

ANALYSIS OF GENETIC OPERATOR VARIATIONS FOR TOURISM ROUTE AND SCHEDULE OPTIMIZATION

Faiz Ihsan Akram, Wayan Firdaus Mahmudy, Agus Wahyu Widodo • Informatics Engineering, Faculty of Computer Science, Universitas Brawijaya
Jurnal Pengembangan Teknologi Informasi dan Ilmu Komputer (J-PTIIK) Vol. 10, No. 6, June 2026

PROBLEM

Daily trip planning is modeled as a TSP with Time Windows plus meal scheduling and travel preferences. Prior studies usually tested a single genetic operator without comprehensively evaluating combinations of crossover, mutation, selection, and repair.

METHOD

80 operator combinations: 5 crossovers (PMX, OX, CX, One-Point, Two-Point) × 4 mutations (Insertion, Reciprocal, Substring, Centre Inverse) × 2 selections (Roulette, Tournament) × with/without repair. Each combination ran 20 times (1,600 runs total), population 100, 750 generations, Pc 0.8, Pm 0.01.

BEST COMBINATION

Insertion – PMX – Roulette – No Repair

The only one of 80 combinations to exceed an overall score of 0.75.

0.7576 Overall Score

1.2875 Mean Fitness

3.74 s Time / Run

→ PMX is the best crossover (mean overall score 0.6534) because it preserves good visit orderings from both parents. Insertion is the best mutation (0.6378) because its changes are local and controlled.
→ Repair raises fitness (1.2529 vs 1.1585), but nearly doubles computation time (7.45 vs 3.77 seconds). As a result, all top-10 combinations run without repair.
→ Sensitivity test: when distance is prioritized (time 0.3 : distance 0.7), Tournament-based combinations climb the ranking, while PMX and Insertion stay on top in both scenarios.
→ Conclusion: the best performance comes from the synergy of an operator combination, not from any single operator. Itinerary quality is not yet consistent across all combinations, especially the breakfast slot, because few restaurants in the dataset open early.

📄 READ THE FULL PAPER

// ENGINEERING PERSPECTIVE

WHY NOT USE AN LLM / DEEP LEARNING?

The Right Tool for the Right Problem: fit-for-purpose engineering vs. chasing hype.

LLM / Generative AI

Poor Fit

✕ Prone to Hallucination: Often gets opening hours and route ordering wrong.
✕ Slow & Wasteful: 3–8 second responses, requires powerful cloud GPU servers.
✕ Expensive: Requires an ongoing third-party API token subscription.

Genetic Algorithm (This Research)

Precise & Fit

✓ Respects Opening Hours: Schedules are checked against opening/closing times from the research dataset, with distances & travel times from a Google Maps matrix.
✓ Super Fast: 25 generations finish in <200 milliseconds.
✓ Zero Cost: Runs entirely in the browser with no extra server.

// THESIS RESEARCH ARCHITECTURE

GENETIC OPERATORS & SPECIFICATIONS

1. INSERTION MUTATION

Moves one location gene to another position without breaking an already good route order.

2. CROSSOVER PMX

Partially-Mapped Crossover with a cycle guard to prevent duplicate genes.

3. TOURNAMENT SELECTION

A tournament (k = 3) picks the best route candidates in a stable, competitive way.

4. REPAIR + PENALTY

Repair fixes chromosomes (post-crossover, opening hours, meal schedule); remaining violations are suppressed by the penalty function. The combination can be configured in Engine Config.

// DISCLAIMER

NOTES & LIMITATIONS

This app is a research prototype for experimenting with genetic algorithm operator combinations, not an official travel planning service.

! Data may be outdated: the list of attractions & restaurants, opening hours, categories, and visit durations were collected during the research. Some may have changed, closed, or be inaccurate.
! Routes are not always optimal: genetic algorithms are stochastic. Results can differ on every run and are not guaranteed to be the best route. Penalties shown in the app flag parts of the schedule that are not yet ideal.
! Travel times are estimates: distances & travel times come from Google Maps at data collection time and do not account for traffic, weather, or road conditions.

→ Always double-check opening hours and site conditions before you go.

TRY THE MALANG TRIP PLANNER NOW

Interactive Leaflet OpenStreetMap visualization, local food filters, and an instant chronological itinerary.

OPEN TRIP PLANNER

MALANG TOURISM ROUTE OPTIMIZATION. GENETIC ALGORITHM, NOT AN LLM.